Thoughts on Pelicot mass rape case: the reason I do not understand it at all, is that men are competitive. And seducing women is an achievement, a trophy, raping unconscious women is not, it does not give one more bragging rights than an inflatable sex doll - rather a definite "desperate loser" vibe.
So I don't get it. Definitely not simply patriarchy, because the competitive attitude to sex is also patriarchy. Perhaps a particularly vile, hateful subset of patriarchy.
Epstein rapists are somewhat more relatable, because getting into that elite exclusive circles, a secret society, is an achievement. But ultimately I still don't understand that, because one can just party on that island without raping teenagers and it is the same achievement. Also, getting into a secret society does not count as a big achievement by the likes of Clinton, Trump or Chopra, those people can get in everywhere.
I think you can't explain urges rationally; it's like asking why people eat sugar when they know that it makes them fat, sick, and unattractive.
Similarly, if someone feels a sexual urge and there is an opportunity... and it is a powerful person who is not used to getting "no" for an answer... and it's guaranteed by other powerful people to be unpunished...
And there is no dilemma. You can rape teenagers from poor families on Epstein's island on Monday, *and* seduce the ladies on Tuesday. (And the dark truth is that most of the ladies won't even mind, because they don't perceive the poor teenagers as their equals.)
Men are competitive, but sometimes they also share with friends. And some of them have weird fetishes. The Pelicot case seems to be about a fetish, the Epstein case is more mainstream.
Of course you can explain urges rationally. How thirsty do you need to get before you're willing to drink -- "insert unpalatable beverage here"? The Male Sex Drive, left unslaked, is a higher strength urge than thirst, which is already a 9/10 (Male sex drive being a 9.5/10).
Explaining Epstein as "urges that don't get punished" is like trying to tell me that f*cking goats is also an "urge that doesn't get punished." It's a poor model. You'd do better conflating Epstein's island with noble cannibalism.
How could you possibly understand it? You are clearly not the sort of person who is a rapist. If you did truly understand it, you would almost certainly feel ashamed of yourself. That said, it’s a pretty strange case.
Intelligent people build models of others, that enhance understanding. If you aren't doing such, your life is probably in danger, due to sheer ignorance (not, in particular, from rapists, necessarily.)
Understanding someone/something is being able to predict it, in novel situations. Either you understand electricity or you do not (yes, there are gradations, not understanding QED might mean you can't do certain things, but you can certainly still plug in a lightbulb).
Other people do not necessarily emulate the laws of physics and chemistry. You cannot compare understanding electricity to understanding other human being.
Emulation isn't necessary to model other humans. In fact, it may be counterproductive, since emulation means you're "thinking like them" and thus you don't need to learn the models from scratch. Learning to build models of humans is a difficult, lengthy process.
I am not surprised by the part where Fox News lies; obviously. But I wouldn't predict that Americans will show someone else's weapons and pretend that they are theirs. That's something I would expect from someone like North Korea, who *wants* to be seen as powerful, but *isn't*.
They played the wrong "drone footage" clip – the guy who was talking, Brett Velicovic, was describing the strikes against Iran's leadership, while the videos were of Iranian drones being hit – and fixed it in a later segment, with rather effusive praise for the Ukrainian technology. Even for Fox News critics, you're scraping the bottom of the barrel here.
Murray Gell-Mann was right, don't trust the media, it never knows what it's talking about. Obvious mistake, possible misleading by government actors, actual treachery, it doesn't matter. These guys aren't correct 10% of the time, and when they are correct, you could have observed it walking down the street.
Current "hysteria in media" for example is about oil, which is like screaming at the slowly moving elephant for being big and gray, while the cheetah is actually mauling your child. Expecting the mainstream media* to actually do their job is a step too far, apparently.
*as opposed to trade publications, who publish Actual News because it's the only thing that's within their remit.
So I am posting my small problem amidst these discussions of world events and big issues: I would like to be able to "Like" posts here. GPT said to install TamperMonkey in Chrome, and then install a script that it wrote. Did that, but the script is not working. When I come here, I still don't get an option to heart a post. GPT thinks the script is OK, and probably I need info from someone here about what works with the present Substack to unhide the hearts option. Can someone point me in the right direction?
FYI, I don't like getting the rare 'like' for a comment. If you like what I say, just write that as a comment and say a little about why you like it. (Or don't like it...)
The past consensus has been not to enable likes, at least as a count display. I agree. I think they encourage group think and people tailoring comments to specifically attract likes, rather than discussion or information, clever insults and wicked burns ensue. Leave Reddit to Reddit.
Doesn’t the Substack app allow people to send a ‘silent’ like to a commenter? I don’t use the app myself but occasionally get a ‘like’ notification.
There is a tradeoff. On one hand, voting may encourage group think, in the bad sense of the word. On the other hand, voting allows people do get rid of spam, or users that were already banned but created a new account, without having to wait for a moderator.
That said, the Facebook-style voting, when there are upvotes but no downvotes, is probably the worst of both worlds, because you can't really use it to discourage spam and low-quality content.
I think Reddit with double votes has more of a hivemind than Facebook. I am kinda undecided about Facebook. Too many anti-science types, which is bad, but people with unpopular opinions flourish there, which is good. That is because hundreds of people can upvote a flat Earth stuff and no one can downvote it. In this case it is bad, but on the whole it is good to not have too much of a hivemind. Sometimes the accepted wisdom is wrong.
I was thinking about this a lot. Ultimately I would like a society with more diversity of opinion, even if it means putting up with cranks and fakenews and Russian propaganda bots.
Because now actual experts like Mearsheimer are not listened to if their opinion is different. Or Jeffrey Sachs. He was always a Russia expert, and he was always listened to in the White House, except during the Biden admin (and I think today with Trump, too)
I agree! I was reluctant to do it at first. I think it had something to do with that indignation dynamic that the Twitter algorithm exploited: There’s something about being infuriated by someone that kept me coming back from more. But once I blocked someone the indignant itch I’d been left with by their final post faded, and the silence was so sweet.
All that gets enabled by the script is a silent like. The person whose comment you liked is notified, but you cannot view how many likes anyone’s posts have gotten.
This post came out too late to be part of the recent discussion on crime statistics and the perception of the same, but it would've been invaluable in providing vocabulary ("felt safety") and hard numbers on the criteria people use to evaluate their safety.
If there's ever a "Highlights from the comments on" thread, I hope I catch it in time to get this into the comments, because that's what most people ended up trying to talk about without really being able to articulate themselves or back up their anecdotes with data.
From the satellite image, Iran consists of a tiny strip of lush green on the Caspian coast, combined with a big swath of arid mountains which take up the other 95% of the country. So you'd expect everyone to live in the green part, right? Well no, as far as I can tell the green part is barely more densely populated than the rest of the country, and all the biggest cities are in the arid mountain part.
Now all the oil is in the south, so maybe that's part of the answer. But Persia has been the leading regional power for most of the past 2500 years. Apparently it has had 32 different capital cities over that time, and I'm not going to look up the locations of all of them but the ones I've heard of like Persepolis, Susa, and Ecbatana are all in the arid mountain bit.
How did a bunch of arid mountains come to support 100 million people and to spawn multiple regionally-dominant empires? And why didn't they all just move to the Caspian coast where it looks much nicer?
It's all path dependency, in an alternate Iran with a Latin American style leadership they'd have moved the capital from Tehran some time in the 20th Century.
I gave a similar question to ai a while back. I thought tehran was on the wrong side of the mountains and the gulf coast was grossly underdeveloped relative to its neighbors.
Basic gist is that tehran is a well defended mountain fortress located near strategic cross roads. From this location its power projects south, east, and westwards. Hence ancient empire stuff. It gets enough water off the mountains for mountain fortress purposes. But it struggles for major metropolis purposes. In modern times all power is concentrated in tehran so they build port infrastructure, but not cities along gulf, partially for defense reasons. They do have an important city in ahvaz a bit upriver from the coast. Expensive ground transportation costs are not immaterial for them.
If you look closely at a good population density map the Caspian coast is the most densely populated area outside of the Tehran metro.
Also, one of the things that we learned in Afghanistan is that a lot of people like herding sheep and goats. They weren't moved by promises of a more modern lifestyle. Success to them just meant more sheep.
Fog of War stuff. Unfortunately, so much has been going on, that I haven't bookmarked everything I should. Here are some interesting tidbits to consider that may or may not be true.
IRAN:
1. Is Mojtaba Khamenei really in charge now? He was the chosen successor to his father, and the Assembly of Experts was meeting to elect him Supreme Leader, thereby destroying the quorum required to make his selection legal. Supposedly, the surviving members met by phone the next day, and under IRG pressure, elected him. But rumors were that many clerics didn't like being strong-armed by the IRG, and they questioned whether he had the clerical credentials to be Supreme Leader. And there are constitutional questions about whether enough members of the Assembly of Experts remain to make the election valid. AFAIK, Western and Middle Eastern media seem to be hedging their bets and are wording it as Mojtaba Khamenei is "set to be chosen as Supreme Leader." Of course, the Israelis took out the offices Iranian TV network in Tehran, so if he were officially elected, the Assembly may have difficulty getting the announcement out.
2. The entire IRG senior command seems to have been taken out (according to Israeli sources). Regional missile commands seem to be functioning independently, which may be the reason we've seen missiles and drones being launched at every surrounding country except Afghanistan, forcing Gulf States, which would have preferred to stay on the sidelines, to react to the attacks.
3. But missile and drone attacks have been falling day by day. Quoting the X post below, "Clear downward trend: Peak saturation early (Day 1-2), then sharp drop as U.S. | Israeli strikes destroy missiles, TELs, storage depots, other launch infrastructure and crews."
And there is some question about whether they have many more stockpiled, having sold the majority of their stock to Putin.
4. Why hasn't Trump pushed for Reza Pahlavi's return? He has a sizeable network of followers in Iran, and he's posted a sample constitution and reconstruction plan for Iran. Pahlavi would become a constitutional monarch and lead the transitional government under his plan. Does Pahlavi not have enough resources to bribe Trump to support him? And Trump claimed their preferred successor (unnamed) was killed in the attacks. Hmmm.
CHINA:
5. Western media and analysts haven't really absorbed the fact that Xi's removal of two of the three top leaders of the PLA had functionally decapitated the PLA military command. Generals Zhang Youxia & Liu Zhenli were removed (arrested? killed?) back in January. This only leaves a political apparatchik appointed by Xi at the top level of the PLA. Xi also purged their immediate subordinates, and State Media said that regional military leaders were instructed to remain on base and attend political training classes. This didn't happen, and various military commands announced maneuvers (with lots of videos being posted on X and other social media of troops marching in formation). Things have quieted down, but...
6. ...but the Political Bureau of the CPC hasn't endorsed Xi's autogolpe (top-down coup). They declined to officially remove Zhang Youxia & Liu Zhenli from their leadership roles at a Politburo meeting at the beginning of February and a second meeting last week. Under the operating framework of the CPC removal of disgraced officials from their positions is necessary before there can be formal trials. Zhang Youxia & Liu Zhenli are suspected to be dead. But there should be an announcement of support for Xi, and that hasn't happened.
7. Rumor is that the Xi faction has pissed off the Old Guard, plus a group of what they call the Reformers. Xi had already purged a bunch of Reformers from the PLA. Although Zhang Youxia & Liu Zhenli got their promotions from Xi, the Old Guard had come to regard them as their own. And notably, Zhang had been cautioning Xi that the PLA wasn't prepared to retake Taiwan.
8. And at last week's meeting of the Political Bureau of the CPC, they struck from the rolls a bunch of PLA regional command officers in the Southeast Region who were perceived as Xi supporters, and would have been the ones coordinating any invasion of Taiwan.
9. Around the same time, state media let it be known that Xi would be OK with the Taiwan status quo if the US didn't sell them advanced military systems. Also, Xi was supposed to host Trump in Beijing for a summit at the end of this month. He invited Trump soon after the initial meeting of the Political Bureau failed to endorse his autogolpe of generals Zhang & Liu, probably to boost his perception as a leader with capable international chops. Chinese and US representatives were supposed to meet in one of the Gulf States to work out the arrangements for Trump's State visit, but the Iran strikes and instability in the Gulf region have thrown those plans out the window. It will be interesting to see whether Xi continues to extend the invitation, because it could be perceived by Chinese hardliners as kowtowing to Trump after the Iran fiasco. And if he cancels, the Chinese perception is that American hardliners will perceive it as weakness on Xi's part. This is a no-win situation for Xi.
10. Reportedly, the Chinese were taken off guard by the massive strike on Iran. Their conventional wisdom was that the US would at most try for the nuclear processing center again. Planners weren't prepared for the US moves. Now they have to deal with the fact that Iran's oil isn't flowing to China; it could (will?) disrupt their economy. And the Iran fiasco has proved that Chinese weapons systems, which they sold to Iran, can't stand up to US weapons. The old guard & the reformers are very concerned. And Xi may not make it to the 21st Party Congress in 2027. Even is Xi wins his political battles, he won't be able to move against Taiwan in the foreseeable future.
RUSSIA/UKRAINE:
11. Russia has lost a major source of drones with Iran being taken offline. Under Ukrainian bombardments petroleum infrastructure in Western Russia, Russia's oil revenues have declined precipitously. Rumor (repeat, rumor!) has it that Putin is considering turning people's bank accounts into war bonds, and that he's considering universal conscription. Hardliners are calling Ukraine a distraction and urging him to go after the Baltics. Especially since they see the US is focused on Iran. It would take Putin months to call up 1 to 2 million more soldiers, and there are questions about whether they have enough weapons left in their stockpiles to arm them.
12. And Secretary of War Hegseth has reached out to Ukraine to assist with US drone defenses in the US Gulf (because our traditional missile defense systems don't seem to be working very well against Iranian drones). Zelenskyy responded favorably to the request, but he made it "clear Ukraine would help only on the condition that its own defence was not weakened and that there were diplomatic gains for Kyiv - suggesting, in particular, that Ukraine would be willing to swap its interceptor drones for more US Patriot air defences to protect against Russian ballistic missiles." (according to the BBC).
Well, we've been blessed to be living interesting times.
5-10) What you outline is good news for the West and bad news for China. Western analysts are aware (at least the smart ones, who are NOT talking to the media, being too busy with a war).
12) Ukraine claims their drones work, quite well. Now, they get a field test, so we'll see.
Last Wednesday was a big bombing day... (Remember, America wasn't ready for this one, they wanted a second aircraft carrier in the region, and Israel saw an opportunity and jumped the gun).
Not any kind of international relations expert, but when it comes to China, I am thinking of Napoleon's maxim not to interrupt your enemy when he is making a mistake. If the US is getting bogged down in a Middle Eastern war and using up all its best weapons and interceptors, just let it happen and keep your powder dry.
Yes. Although Iran has purchased a lot of Russian milware, Iran started buying from the Chinese, who offered a better price point. And the Chinese equipment seemed superior to the Russian equipment — it's an open secret that the Chinese copied US weapon systems whole-cloth. But there's corruption in China's military procurement systems. Maybe not as massive as Russia, but some analysts have pointed out that generals and officials have been cutting corners on QA and software updates and pocketing the difference. Pakistan also bought some Chinese weapon systems, but they failed miserably in a recent spat with India.
According to ChatGPT, Iran bought a bunch of China's HQ-9B long-range surface-to-air missile systems (and a bunch of other radar defense systems) to protect critical sites, but reportedly failed to effectively stop Israeli and U.S. strikes.
Possibly. But Ukraine has shown that RU missile defenses can be overwhelmed by drone swarms and drone misdirection. They can only fire some many expensive anti-missiles missiles before they can't respond any further and cheap drones overwhelm them. And UKR is using new cheaper interceptor drones for the slower stuff that comes over from RU.
It may be that the IRG are just missile defense dunces.I don't know how the US and Israel are taking out their setups, but I'm pretty sure it's not with drone swarms (because neither country has developed its drone warfare). But Pakistan had the same problem with same type of equipment Iran used, taking down Indian missiles when India struck some terrorist camps inside Pakistan.
Likewise, Iran is using relatively cheap Shahed drones to hit US targets and civilian targets in the Gulf, and the US hasn't been able to stop them all. So, I suspect that we'll be arguing about this with 20-20 hindsight for years to come. :-)
Both. The need for an "in person meeting" cost Iran dearly. Their dead-men switches were designed for an all-out assault, and that's not what the Saturday Massacre was. Hence the rest of the middle east now having a fig leaf to attack the heretical ummah (under the self-defense provision of Islam).
2. ”missiles and drones being launched at every surrounding country except Afghanistan”
Including Turkey? I read that Turkey had not been attacked, perhaps because of the ramifications of a direct attack on a NATO member. Is that old news?
4. “Why hasn't Trump pushed for Reza Pahlavi's return?”
It has been suggested that Trump’s reticence about long-term goals is intentional, to prevent such goals from turning into requirements. I think most of American, MAGA or not, has a pretty jaundiced view of “nation-building”.
More realistically, the goal is nation-destroying. Iran lashing out at the world in a final, desperate attempt for survival is only going to encourage everyone to finish the job.
Iran is literally surrounded by enemies and in that light, it's broad choice of targets has been fully understandable. Perhaps not the wisest choice, but an empathetic and understandable one. I don't think people fully realized before now just how deeply enmeshed Iran's neighbors are in America's broader military system, whether officially or de facto.
Of course, part of the reason why Iran has all these enemies is because they have insisted on exercising power through clients, and have refused to recognize the principle that one country shouldn't meddle in the internal affairs of another.
But however we got to this point, it's obvious now that Iran has an absolute strategic need for nuclear weapons. Despite what you may have heard, the nuclear program was subject to fierce debate in Iranian society. I am sure that is over now. I will bet you that every iranian, from the monarchists to the regime stalwarts, now recognizes that attaining nuclear weapons is the only safety for their country. If they had a consensus on this before, they probably would have built those weapons long ago.
As far as "destroying" Iran goes, that is a totally fanciful war goal. Even after two nuclear strikes, Japan was not destroyed as a nation. And there is no stomach in America to subject Iran to even that level of bombing. Or to allow Israel free reign to do that.
The deadman switches were conceived with an "America goes full boar" mindset.
They weren't conceived with "surgical Israeli Strike" in mind.
So, the Iranians look foolish, and bellicose towards their neighbors -- the idea of their strikes was to get their neighbors to lean on the United States, but in practice, it's triggered their "self-defense" clause (no attack muslims without self-defense being involved).
Iranian nuclear program has been DOA for years upon years. Assassinating Americans is what got Iran in trouble, and nuclear weapons don't fix stupidity.
No, we will not "destroy Iran" as a country. There will still be an Iran, or something like it. Ukraine will disappear before Iran does.
Al Jazeera reported that as of March 4, 2026, Turkey's defense ministry confirmed that NATO air defenses intercepted and destroyed an Iranian ballistic missile (and in some reports, drones) headed toward Turkish airspace. No casualties or injuries were reported.
And for some reason, the Iranians lobbed some missiles at Azerbaijan. But AFAIK, they haven't lobbed any missiles at Pakistan since January of 2024. Not sure why. I thought relations with Pakistan were kinda tense.
Surely thats why. They attack countries that dont want to get involved and might pressure the US to stop. If they attacked pakistan theyd be inviting serious counter attack they are in no position to deal with
Despite supportive remarks from Azerbaijan to the Iranians at the opening of Operation Epic Fury, Iran attacked the main Azerbaijan airport to disrupt international air traffic that was now passing over Azerbaijan, which had been diverted from International air routes south of the Gulf. They also targeted an Azerbaijani elementary school. The drone missed the school, but you can hear kids shrieking in the video. Iran has denied attacking Azerbaijan, but President Aliyev has announced that Azerbaijan will respond militarily to the unprovoked attack. They've canceled all military leaves and announced a state of emergency. Don't know if their weaponry can reach far over the Iranian border, but their armed forces have combat experience (in their war against Armenia). It's worth noting that the northwest corner of Iran is populated by ethnic Azeris who carry a grudge about their treatment under the mullahs. An independence-minded Azeri group has just asked Azerbaijan for help against the central government.
Iran tried to hit the Georgian and Turkish links in the pipeline that goes from Turkmenistan under the Caspian Sea to Baku, down to Turkey, from which Israel gets a lot of its petroleum and gas. I guess they missed, and it's still running.
Dead-men switches don't respond to "current intelligence." they respond to projected future intelligence, based on when the plans were drawn up.
(And someone may have wanted to spoil some Israelis vacations in Cyprus. When you're to the point of using dead-men switches (aka preprogrammed orders), the artillery officer gets to choose. "Yes, No, Point Somewhere Else"
The average commenter here seems deeply jingoistic and in favor of an aggressive foreign policy against any country perceived as an enemy of the West. With their very technical interests, I imagine the average commenter doesn't have much time to study these things, and so when it comes to foreign affairs, they are left to passively sign off on the manic aggression of mainstream corporate media analysis. This is the last place I would come to learn about world events and foreign policy.
I would argue that anyone who isn't jaundiced and cynical about this Iran operation is regrettably jingoistic and militaristic.
Hell, anyone who defaults to a neutral, "let's wait and see" attitude about American military operations is passively jingoistic, and is likely naive about America's post-1945 military legacy.
Are you in favor of propping Xi up with military action? Do you see this as an aggressive foreign policy, against a state that believes Trump is an existential threat?
America's foreign policy "ideology" is quite a bit different than you'd think. Obama put it best, "America doesn't have allies, only interests."
If it's in our interest to prop up Xi, we will. If it's in our interest to assassinate Zelensky, we will. Etc. America goes around "color revolutioning" many countries, and I'm sure you cheered for the Arab Spring, which was a very aggressive foreign policy decision that has paid off extraordinarily poorly.
Then I'm going to count you as one of the few principled...isolationists? Something like that*. Few people can stomach "Obama was cuckoo-crazy too" (let alone Biden, who had former military Democrats saying "he's trying to start World War III" -- scuttlebutt supports this, alongside his appointment of neocons to run Ukrainian policy.)
*wanna call yourself a "chary interventionist"? or just "not a chicken-hawk"?
As a confirmed peacenik, our Obama/Clinton/Biden-era Ukrainian policy has me pretty ticked off.
I'm also a peacenik. I only believe in using military force as a last resort. Since 1945, I can't think of a single American military intervention that was honorable or necessary. Perhaps there was some obscure operation that was worthwhile, but certainly none of the headline grabbing operations were reasonable.
Funny, I read hella lot of stuff every day. Happy to do the doom scrolling for you. For whatever reason, it doesn't upset me as it does my fellow Lefties.
And as long as Trump doesn't put US boots on the ground, I'm OK with Operation Epstein Fury. Fuck the Mullahs. They didn't deal with Obama honestly when he tried to get them to scale back their nuclear ambitions. Best to take them out while we still can.
Bibi's been saying the Iranians are "very close to a nuclear bomb" since the early 1990s. "Scale back their nuclear ambitions"? I don't think they HAVE nuclear ambitions, to be frank and honest (there's good recorded evidence from around Bush's Axis of Evil that the Iranians didn't want a bomb if it would mean an invasion). I'm not sure our military assessment says so either.
That said, putting assassins on American soil hunting Trump and other military was a very bad ploy, and worth toppling their government over.
Quick note on point 11: Russian Shahed production is almost entirely domestic, with a primarily Chinese (not Iranian) supply chain. I don’t expect the war with Iran to have any significant impact on Russian drone production.
My understanding was that many of the electronic and mechanical components were being imported from Iran. Also, Iran had sanctions-avoidance network in place for money laundering and electronics imports, so RU has probably lost this for at least now. And now that UKR has taken out two Sahed manufacturing facilities with their flamingos, we'll see if RU can keep up their manufacturing pace. Also, a bunch of electronics manufacturing facilities were taken out pre-Flamingo. I don't know if RU has gotten them completely back online. But RU is still getting German electronics for their radar and anti-aircraft systems. And China is providing electronics for RU's smaller non-Sahed drones. There was a good write-up on this last year. Sorry, I didn't bookmark the site.
My understanding is that the original licensed production run of domestic Shaheds was essentially from Iranian kits, but since then Russia has modified the design into a number of different variants that haven’t been shared with Iran. They’re definitely using a significant quantity of western-sourced parts, although my understanding is the majority (maybe just a plurality?) of them are coming through China and few if any are coming through Iran.
What you say is likely true. However, Chinese electronic components, including Western-branded semiconductors, reach Russia primarily via Hong Kong middlemen, then on to Iran, Turkey, and some Central Asian countries. Chinese manufacturers want plausible deniability that they're not avoiding sanctions. HK is a big transhipment point for US technology into sanctioned countries.
Regarding 4: There is a big difference between the Iranian population being upset at their government, and the Iranian population wanting their government to be replaced by an American puppet. You’ve been reading too much CIA propaganda if you think Pahlavi is viable barring a full invasion.
All those protestors were waving the old regime's Lion and Sun flag. Seemed like there was a lot of popular sentiment for the Pahlavi on the ground. And your understanding of the US deep state is faulty. The CIA doesn't have a propaganda arm anymore. The State Department does, but since Trump zeroed out their foreign-language communications budget, they're not putting out much propaganda material anymore. As for internal US propaganda, that's handled by Fox News.
As for putting U.S. boots on the ground in Iran, I'm sure that would drive popular opinion the other way. And let's face it, the Pentagon can stir up the hornet's nest, but they don't know how to deal with asymmetric warfare (and one could argue that they haven't fared well in any long-term ground war since WWII). Let Pahlavi return. It's the best realpolitik solution.
The thing those protesters were conspicuously *not* waving, is likely to be decisive. They were not waving guns.
Almost everyone in Iran with an actual gun, is solidly on Team No Pahlavis. If they weren't, they would not have been allowed to have guns. OK, there's probably a some people in the regular army who'd be OK with Pahlavi and whose technical skills made them un-purgeable by the IRGC, but not enough to matter. And there are some armed Kurdish insurgents, but they're doing their own thing, and there are maybe some resistance groups that stole some police guns a month or so back, but again not enough to matter.
What you're seeing under the Lion and Sun flag was the minority of Iranians who are largely urban, educated, English-speaking, and unarmed. What they want doesn't matter now, because all the people with guns are looking for *someone* to shoot and they can't shoot Americans or Israelis because an AK-47 can't reach 30,000 feet. They're going to be keeping their heads down, or they're going to be dead.
If you want a Pahlavi regime, that's going to take American boots on the ground. Either a *lot* of them, or a lesser number of special-forces advisors working to build up an effective Iranian resistance over many years. I'm pretty sure Trump isn't up for either of those.
According to some news reports, the IRG finally had to bring in a few thousand paramilitary groups from their network of sympathetic terrorist groups to put down the protestors. During the protests (and before Operation Epstein Fury kicked off) there were explosions at IRGC facilities and bases, fires at missile depots, attacks on military warehouses, and sabotage of communications or power systems. Khomeini (the elder) went into hiding, and the IRG was kicked back on its heels until they received reinforcements from their external assets to put down the insurrection. Some have blamed Mossad for the attacks, but Israel has remained silent on them. And some have suggested that the attacks were carried out by Mossad-trained Iranian dissidents. Israeli intelligence has been working with impunity in Iran for a while. Last year, Mahmoud Ahmadinejad admitted that an Iranian counter-intelligence team tasked with locating Israeli spy cells was led by a general who turned out to be working for the Israelis. And someone tipped off the Israelis precisely when and where Khomeini convened a high-level meeting of his cabinet and his advisors, decapitating the senior levels of government.
The protests in Tehran kicked off after a major scandal in which Tehran's water supply was diverted to agribusinesses controlled by senior government officials. Tehran was running out of water, and Iranian President Masoud Pezeshkian stated that moving the capital from Tehran to the south was "no longer optional" and is essential due to severe water shortages, over-expansion, and land subsidence. But it was clear the agribusinesses were getting all the water. Moreover, government contractors were charging exorbitant prices for the water they trucked into Tehran. This, on top of rampant inflation and job losses, pushed Iranians out into the streets. I suspect that most would be happy to get rid of the mullahs.
Israel's responsible for some part of the water shortage, due to damming up rivers upstream in Afghanistan. I'm willing to call vanishing the water supply a war crime (as I've done when Ukraine did it to Crimea) -- it's punishing civilians.
Protests are one thing (Iran is often protesting), but explosions et alia sound an awful lot like CIA/Mossad (maybe our own milint if you really want to say CIA is out of the game -- which I maintain they're not).
The whole 21th century Middle Eastern experience seems to indicate that local "puppets" have a lot more agency that Americans think. This is a region where political scheming was perfected long before Columbus sailed the Western ocean, and I absolutely wouldn't be sure of who is a de-facto client of whom.
Or 20th century experience too. Saddam Hussein in Iraq, the Taliban in Afghanistan were supposed to be anti-Communist "American puppets".
Indeed the entire Islamism, which caused a lot of terrorist trouble in the West, seems largely a Western creation. Back then Brits encouraged Wahhabism, Salafism, as the Wahhabites really hated the corrupted and nearly atheist Turks. Then America encouraged them further because they really hated the atheist Communists. The Taliban got a massive number of Stingers when they were fighting the Soviets. That really mattered, because as they said "we are not afraid of the Russians, but we are afraid of their helicopters".
I heard that the rural parts of the country are more religious and much less West leaning than the population in the cities. But I don't know how to check that.
- The "have someone more pliant from inside the regime replace the old guy" seems to have worked out for Trump in Venezuela, so he's trying to repeat the model (for one thing it's very much not the Iraq 2003 style total upheaval nation-building project, which Trump admin is very insistent it's *not* trying to repeate)
- related to the above, Reza Pahlavi just plain gives rather strong Ahmed Chalabi vibes
- this whole "combat operation" still feels like they're flying by the seat of their pants, which is of course Trump's general MO for most things (it's got him this far!), and strong commitment to *any* option would not facilitate this model of operations
Trump seems to be going for "let's bomb the hardliners and leave the moderates alone" with a strategy that consists of: "continue to bomb until someone starts crying Peace!"
Venezuela was a specific "take Venezuela free oil off the board for Cuba." He didn't need much buy-in from the next Venezeulan dictator (and the "we own you, wear what we say" was kinda explicit).
Moreover, if Exxon/Shell/Chevron wants its oil facilities back online, they're going to have to spend tens or hundreds of billions of dollars to rebuild the facilities that rotted under the Chavenista's neglect. I'm sure Delcy Rodríguez would settle for a smaller piece of a much bigger pie.
My closest Persian friend, an American citizen for at least two decades but with family in Iran, was disappointed that the Fordow attack last year was so limited. “Why did they stop?” he asked me.
But he is disappointed by the “cut off the head” strategy, which struck *me* as brilliant and measured, like Israel’s pager attack. He says the “Muslim Mafia” is too big and deep for this to work and fears we will eventually make peace with some low-level Mullah, every bit as fanatical as Khamenei but perhaps more deceptive, and nothing will have been accomplished.
In the interest of harmony I didn’t ask him what he thought we should do. I understand that what’s good for America and what’s good for Iran may not be identical.
One guy was invited to a Persian restaurant by an Iranian friend, then the door opened and 8 incredibly arrogant guys walked in and the Iranian friend started cussing them out. They were the consulate people and everybody around the tables hated them. The point is a real revolution would mean a massacre of elites and they know it.
After the Fascist Grand Council fired Mussolini, they asked the king to find a successor, who found Badoglio, a largely non-political soldier who installed a government of bureaucrats and soldiers and the Fascist Party was disbanded. He had an impossible job, to end the war without being invaded by Germany. So he tried secret negotiations and it did not work out too well.
That only happened because we first completely destroyed Germany's armed forces and industries, conquered and occupied most of Germany and made it clear that we were taking the rest in a matter of weeks at most, and killed several million Germans in the process. Absent that, Hitler wouldn't have killed himself and if he had been killed, whoever succeeded him would not have surrendered.
Even the Stauffenberg variant of Operation Valkyrie would not have resulted in (immediate) unconditional surrender.
The United States may have mostly destroyed Iran's Navy and Air Force, but it is exceedingly unlikely that it will destroy the Iranian Army (or the IRGC's land forces). There has been very little in the way of industrial destruction that I know of, and a policy of wholesale deindustrialization seems also unlikely. The idea of American armored divisions marching on Tehran seems frankly absurd. And while it's possible Trump is willing to kill several million Iranians, that would be difficult and tedious to do without nuclear weapons.
So, yeah, meet the new Ayatollah, same as the old Ayatollah. Even if we play multiple iterations of that game, we'll keep getting the same outcome.
You need to be wary about Iranian friends in the west. They have fled the regime and are antagonistic to it, and are often antagonistic to the entire country. Like White Russians who were extremely anti communist and often anti Russian. It’s probably not that popular in Iran to support either Israel or the US, I’m sure the bombing the children’s school and the recent mass killings in Palestine don’t help either.
Iran is actually a republic. They have a democratic process. It’s flawed but it’s there. Replacing the top guy is all very well but it’s still replacing a dictator (of sorts) with another. That happened before and it didn’t end well. Since there are elections and a president it seems to me it’s just a matter of replacing the role of the supreme leader. Is that being discussed or is the Trump regime relying on people’s ignorance of the internals of Iran to convince us that he has a guy.
I can certainly sympathize with that. I presume the reason was that they were waiting for solid intel about the whereabouts of the first targets. It’s unfortunate that the timing worked out so badly.
Further, perhaps more empirical arguments against superintelligence. The Transformer model, the basis of all modern AI, was only intended for machine translation. However they trained it on texts that contain reasoning, and it picked up that pattern. This is delightful so it now gets trained on a lot of philosophy and science texts. But the upper limits of its reasoning are these texts. So 140IQ.
Of course the 140 IQ combined with vast knowledge, very fast thinking and the ability to communicate with millions of people can STILL be totally dangerous. People will be tempted to give it too much power in order to beat the competition or win a war.
So I am not arguing against AI could make humans extinct. Just give it nuke launch codes aka SkyNet and we could all die. Or killbots + automated killbot factories. I am arguing this does not even require superintelligence.
So the Terminator movies were directionally correct. SkyNet was mega powerful because of all these, and nearly made humans extinct. But it was not a superintelligence, humans could still outsmart it frequently enough.
Although I will also say that one thing was not correct, they expected deterministic algorithms, so SkyNet was understood as too rigid, not as creative as humans. This is not so with AI which can get very "random" thus creative depending on where the temperature slider is set.
I don't think you get to 140 IQ with an unthinking LLM unless you can curate the texts and say "these are the ones by the geniuses; weight them much heavier than the rest". And the volume of text required, I think makes that impractical.
The current approach seems likely to asymptotically approach the intelligence of the average internet user, albeit one who has recently done all the reading in the subject at hand. And that's not to be underestimated; cheap fast well-read mediocrity can change a lot of things. But I agree that it's not a path to superintelligence, without adding something currently unknown.
You could also do stuff like ‘train it on the arxiv’ which will asymptote it to the average intelligence of an arxiv poster (probably higher than an average internet poster), albeit in a narrow domain.
I think current model Chat and Claude, when analyzing physics and well prompted, are pretty close to a mediocre professional physicist who has recently read the entire arxiv.
Within my narrow domains of expertise I could give you better instant answers than AI, but step a little outside those boundaries and I could not. Well probably I could but it would take me 5 months instead of 5 seconds.
Yeah, I can see Arxiv having enough good semi-curated material to constitute a reasonable training dataset within its domain.
Well, at least until the LLMs start publishing on Arxiv, then all bets are off. Fall back on peer-reviewed journals, and watch the entertaining legal brouhaha of OpenAI vs Elsevier?
Only in domains where you can get ridiculously large amounts of training data with success and failure being graded automatically and cheaply. If you want say a chess-playing LLM, sure, you can have it play bignum chess games against different instances of itself. The ELO scores speak for themselves, and the math is simple. If you want instead an LLM that can fly an airplane, OK, you can throw bignum instances into flight simulators, but that's a lot more compute and it's going to leave blind spots in those aspects of aviation that aren't well captured by flight simulators.
Most of the interesting problems are going to be in the second category, I think.
IQ isn't conserved via textbook. If it were then it would be impossible to produce a human who was smarter than the smartest previous author. What book taught Einstein to have an IQ of 180?
Einstein never had his IQ measured. That's just a bullshit estimate by Catherine Cox, an American psychologist, and one of Lewis Terman's team studying gifted people. In 1926, she published a study titled _The Early Mental Traits of Three Hundred Geniuses_ which estimated the IQs of famous doers and thinkers. Goethe beat Einstein's estimated IQ with something like a 220 IIRC (which isn't a valid number by today's IQ tests). Cox just made estimates that Einstein had a super IQ because none of the ~1500 geniuses that Terman tracked throughout their lives did much of anything noteworthy.
Einstein himself was modest about his intelligence, saying, "I have no special talents. I am only passionately curious". He further emphasized persistence over raw talent, saying, "It's not that I'm so smart, it's just that I stay with problems longer". I believe he also said that his first wife, Mileva Marić, a mathematician, was smarter than he was, but I can't find any quotes, so I may be misremembering it.
I have no idea what Einstein's IQ was - that was just a random above-average guess. The point is that he was far more intelligent than any of the authors of the textbooks that he read as a student. The best scientists rarely write widely-used texts.
Modesty is poor evidence. All it measures is personality factors. Most famous scientists say that their intelligence is nothing special. The only thing that that is evidence for is that most scientists care about their social status. It's beyond doubt that their intelligence is, in fact, far above average.
>I believe he also said that his first wife, Mileva Marić, a mathematician, was smarter than he was
Yes, men frequently lie about their wives because it makes their personal lives much more comfortable. There is zero evidence that Mileva ever had a scientifically important idea. They separated in 1914. Albert published many important papers after that point. Mileva never published anything and there is no evidence that she ever did anything privately. She never even finished her degree because she wasn't able to pass the final exams (she failed twice). The notion that she contributed in any way to relativity is nothing but revisionist feminist nonsense. She was nothing but a housewife.
Most "famous scientists" KNOW their intelligence is nothing special. That's part of what makes a genius, testing small ideas and fitting them together. Geniuses and "sub-normal intelligence" folks have a lot in common. It's the midwits, who memorize things, that think "they must be Very Smart and Very Special."
Geniuses don't need to be "extraordinary intellects" -- they just need to work harder, more persistently, and develop quality control. Intelligence is something that you can continue to build your entire life.
No, they know that intelligence by itself is useless. Hard work is also essential. Since intelligence is out of their control but hard work isn't, they adopt the attitude that "intelligent isn't important" in order to motivate themselves to do hard work. They're wise enough to know that if they sat around thinking of themselves as brilliant that they'd never do anything.
Great scientists have both extraordinary IQ and extraordinary work ethic. No one of normal intelligence has ever won a STEM Nobel.
"No one of normal intelligence has ever won a STEM Nobel." https://en.wikipedia.org/wiki/Philipp_Lenard was an experimentalist who had almost no understanding of theoretical physics. He was just a relentless experiment-builder.
Interestingly, Nazis (like Lenard) disliked IQ testing, because they thought it measures "Jewish tricks" and they attributed success to hard work, not IQ. Lenard was apparenlty a hard worker and not a good thinker. Edison also attributed his success to work, not IQ, he had no theory of lightbulbs at all.
I don't really know who is right in this. Just wanted to show this perspective. Success through "brute force" is a very typically German thing.
Intelligence is hardly out of a genius' control. If it were, they would have continued to test as an idiot on the IQ tests*. Increasing one's intelligence is done through hard work and better models. And as you reach diminishing returns, you STILL don't feel intelligent -- because one of the fruits of true intelligence is capturing the uncertainty of your knowledge.
*People aren't born intelligent. They grow that way, and it's a painstaking process that children who merely mirror the teacher never truly experience. Hence why "gifted" children so often have middling careers.
I agree about Einstein's wife, but with regards to Einstein's IQ, intelligence is not the only important thing, there are many other factors involved, including pure luck. Therefore, if you choose e.g. 1000 people with IQ above some threshold, the most successful one of them does not necessarily have to be that smartest one, or even one of the smartest ones.
I mean, Einstein definitely *was* super smart. I only object against attempts to estimate it by comparing to achievements of other people. More precisely, high IQ is necessary to invent (or even understand) some things, but is not a guarantee of success, even if you try to do something like that (i.e. ignoring the fact that different people specialize on different things). I can imagine a parallel universe where Einstein had a different boss, or got sick, or had too many problems at home and couldn't focus on research, etc., and in that universe he perhaps didn't invent relativity, but his IQ was the same.
To some degree, we can substitute time, money and headcount for raw intelligence. If 10x130 IQ researchers can write a more useful textbook than 1x140 IQ one, it follows that 10x140 might be able to outperform 1x150. I could believe that this gives diminishing returns, but it doesn’t obviously prevent the creation of higher-IQ text.
Yeah there is a pretty standard ensemble argument from classical machine learning that suggests that a "panel of experts" that pools information can outperform any individual expert. You can achieve this even with simple models, e.g. polynomial regression + random forest + simple one-layer neural net. It does not give you unlimited gains, but seems pretty reasonable that 3 x 150 IQ models gives you >150 IQ. And your "time tradeoff" idea seems right as well - in a lot of ways the difference between a 150 IQ and 120 IQ is moreso "processor speed" - a 120 IQ person is still capable of learning multivariate calculus, it just takes longer. And many of the concerns about powerful AI still apply even if you "merely" have Dario's country of geniuses in a datacenter (100 million 150 IQ people at 10-100x human thinking speed) vs. the Machine God with IQ = 10,000.
I don't think handwaving is quite enough to support the logic. You set out some facts and give some conclusions, but I think extra work needs to be done to show that the conclusions follow from the facts. Specifically, humans are clearly capable of writing down ideas that were never in their training set. Can you articulate why you believe the set of written material ingested imposes some kind of constraint on computation performed in silicon that is not imposed on computation performed in meat?
Personally, I agree that modern AI is limited, however I suggest that the limit is more likely to arise from our choice of training model / loss function than from the transformer model. A network of the latter can be used to build aribtrary systems. It is the combination of the former with the training data that determines what systems actually get built; and I don't believe that the missing piece is the training data because that's all /humans/ have to work with and it doesn't stop us inventing things.
So, what I read in your sentence is that instead of being constrained to do the task it was originally designed for, the software became able to do things its creators did not design or anticipate. This seems to be an argument for the exact opposite of the conclusions you are trying to use it to support.
If you intended my takeaway to be something else, please could you expand on what that is and how it follows from what you've said?
Well my point is that it can horizontally deviate from design, that is, do something else that was designed, but I think vertically not, because if it is not in the training data then not able to. Others pointed our reinforcement learning, yes, if there is a clearly defined goal, such as winning chess, and there is self-training possible, like simulate two strategies against each other. But when we are talking about superintelligence, we generally mean something that can talk humans into giving it control over nukes. Superintelligence - I am not moving goalposts, this is literally why Yudkowsky considers it super dangerous - is primarily about an extremely good communicator who can talk humans into giving it power.
When "the training data" is literally the entire text output of humanity, and this is sufficient for humans to come up with things that humans before them did not and could not, I am not really sure I understand what it is that is "not in the training data" that could be.
Have you considered the effect of the choice of loss function on the result at all? I've been talking about that this whole time - about the fact that it is the loss function that determines what it is we use the training data to actually build - but you've not mentioned it in any responses, so I am left wondering whether there is some reason you are focusing on the training data instead.
>Can you articulate why you believe the set of written material ingested imposes some kind of constraint on computation performed in silicon that is not imposed on computation performed in meat?
Just to hazard a guess here. The only thing that occurred to me is highly random associations that turn out to be useful. And are largely informed by our intimate participation in meat space.
Exactly. In these cases the AI could generate training data, indefinitely, by playing. If we were able to give LLMs a similar environment, maybe we could do the same. I'm sure many labs are working on that.
Google did this with other non LLMs, see deep mind blog on ai agents playing various 3d games.
Ilya on Dwarkesh podcast discusses the difficulties with training on reasoning chains that get too long, however.
Assuming that the constraints of human IQ tests apply to LLMs is nonsensical, for too many reasons than I'm willing to list here. I'm not saying LLMs aren't limited by the intelligence of their training material, which is obviously true. But trying to identify that limit via IQ is...like trying to measure the energy output of an atomic bomb based on the weight of the entire bomb and nothing else. Not a perfect metaphor but whatever. The real answer, for this whole spectrum of questions, is "we don't know."
Edit: a better metaphor might be: building a two-legged humanoid robot and estimating its top speed based on fast humans.
I agree with you that applying IQ to LLMs in this way is somewhat silly; however, I disagree with you when you say that "the real answer ... is 'we don't know'". I think it's obvious that LLMs cannot become "superintelligent" (to the extent that term has any meaning) or even as intelligent as humans (ditto), due to the way their architecture works. Unlike humans, LLMs are e.g. unable to learn on the fly (re-training is incredibly expensive), to fail gracefully when confronted with novel inputs (or occasionally even ordinary inputs), or to synthesize many different inputs (other than text). Some of these weaknesses can be mitigated somewhat (usually by dramatically increasing computing power), but ultimately the problem stems from the underlying NN architecture.
Granted, it should be possible to use LLMs as a component in some other AI architecture to produce an intelligence that thinks more like humans do, but thus far no one had managed to do it, and it doesn't look like it will happen anytime soon.
That said though, a tool does not need to think like a human in order to be useful, and indeed quite powerful: for example, spreadsheets don't think like humans at all, but possess superhuman powers of calculation. And a tool does not need to be superintelligent in order to be dangerous in the wrong hands; for example, bulldozers are really stupid, but can cause a lot of destruction when improperly applied. The more powerful the tool, the greater the danger, and sadly LLMs are quite powerful...
Just in-time for the End of flu season, I came across a podcast from last fall that included a segment on getting your flu vaccine—I've never had issues with mine.
One of the things they say about Influenza is that it can make one susceptible to Bacterial Pneumonia. Isn't that Also a possible escalation for a cold, bronchiolitis, and a crapton of upper respiratory infections that Aren't the flu?
Yes. The teaching in medical school is that are lungs are constantly exposed to germs entering from our mouths (think swallowing down the wrong tube) and from our environment. Inflamed or damaged lung tissue is both a more favorable environment for the growth of bacteria and also less able to fight off the infection. Influenza, as compared to other viruses, is more often associated with severe presentations that are likely to predispose patients to bacterial infections.
> As mentioned above, the US government doesn’t have a formal legal definition of domestic mass surveillance, only “bulk collection.” And the US government has, basically long maintained that even if they hoover up a bunch of information indiscriminately, they haven’t done bulk collection so long as their individual queries against that mass database are more targeted when humans look at them. As a result, at least one Director of National Intelligence said under oath “no” when asked “Does the NSA collect any type of data at all on millions or hundreds of millions of Americans?” even though NSA has admitted it does by the ordinary meaning of this question.
I am very familiar with USGOVT system and know several of the TIA architects. And efforts before TIA (and even back a bit to Operation Ivy Bells). US Govt is very good at say node analysis in a time series to know what's going on, despite the package being encrypted. Walz got caught up in that way with his Signals-encrypted message recently. No warrant required, but useful for influence.
My point has nothing to do with USGOVT. Anthropic, through user prompting, has the equivalent of mass surveillance. They know your concerns, tone, typing speed, voice, etc. They could gain huge knowledge about you, and even use Claude to steer you, and are not subject to the rules put on the USGOVT. Same is true for OpenAI and Gemini. They are unregulated in this realm. Other than market forces.
I can most certainly say the USGOVT has thought about the various ways this can go wrong much longer and deeper than Anthropic, OpenAI, and Gemini.
Anthropic might have a lot of user data, but they aren't buying cellphone tracking data the way the government is. They can't tell that you went to a protest unless you deliberately inform Claude of that fact. The government can.
Anyone can buy cell phone data. Is that what people think is mass surveillance? Anthropic type platform is a non government version of Telescreen. The technology could easily become a vehicle of thought police, whether they are private or USGOVT.
PS: USGOVT can track the movement of every cell phone around the world without buying the data. Likely has been able to do that since cell phones were created. They don't need anthropic to do that or analyze it. Anthropic would have helped when Tia was built, but that is old tech. USGOVT buy data for Parallel Construction.
Go to a motel, and get your license plate captured (this happens every day, to every motel in America). Mass Surveillance. (This is a Private Eye network of folks, not governmental).
The US needs to obtain a warrant if it seeks to prosecute. They don't need to do it for a foreign person. But that is not my point.
Anthropic has been worried about USGOVT. But Anthropoc can, synthetically, gather a lot of invasive information through your prompting. So they are de facto a mass surveillance data. They also have the means to steer you in their response. Likely a far more serious violation of individual rights than how NSA operates CONUS.
People miss that you can synthetically replicate private data from more public data.
At least with ChatGPT, Users can choose to have off-the-record conversations which are not stored / trained on by default. I presume a court would need to provide a wiretapping warrant to obtain those. As ever, it remains bad opsec to produce electronic records detailing illegal acts.
Transformers are not the path to ASI or even AGI. A transformer which has the global optimal parameters to minimize the loss for the standard internet corpus out of the box, is not going to be AGI
Transformers have nothing to do with what training objective function or data is used.
Even if it did, LLMs are also not minimising the loss of internet corpuses since ~GPT-3 era. We have _at least_ 2 generations of breakthroughs in that time (RLHF + RL on CoT).
Because human beings are still better at navigating the semantic ocean. seriously. I have come across this with Claude on many occasions and we have discussed it. I posted it a text by Franz Kafka once. I asked what it thought the most striking thing about the story was. It came up with a very concrete thing that was observable; something to do with the transition from the realm of thought to the realm of physicality. I came up with a different one. The one I came up with was knitted semantically across many domains that we had discussed in detail and embraced things that it knows about our previous discussions. (I say this so certainly because in other conversations, it has referenced those things so it’s clearly included in it’s general memory of dealing with me.)
I doubt those sorts of things can arise in a second order derivative of living-in-the-world.
Will there ever be an AI that can pass a Turing test of convincing you that you are in a conversation with Robin Williams?
I just realized superintelligent AI is not possible and it had to be obvious all along. First of all, can we say the model is intelligence and training is knowledge? So the model itself is a traditional software that is designed traditionally, right? So the question is, how could you design something smarter than yourself? Neither a human nor self-coding AI can do that because you just cannot imagine what it is like, what it is does. Can a 8 years old design an adult? Yes, he can give us the specs of his ideal dad. But we adults would not necessarily agree with that spec. Besides, there are other aspects of being adult. The 8 years old cannot design the thinking process of a 30 years old programmer. Self-coding AI cannot do that design either.
Or you can argue that intelligence is not only in the model design, but in the training too. In that case just where can you find examples of superhuman knowledge to train it on?
On the other hand, AI does not need to be superhuman to be really powerful. 140 IQ with immense amounts of knowledge and very fast thinking speed and ability to communicate with millions of people simulateously would be very powerful.
Humans can absolutely design a smarter than yourself person. Autists do it all the time, because they don't start with the basic inbuilts. When you get used to building, say, facial recognition yourself (or emotional-reading of faces), you can learn to build a lot into your brain.
> First of all, can we say the model is intelligence and training is knowledge?
No, we really can't. You can't separate "intelligence" and "knowledge" in LLMs, they're fundamentally entwined. This is decidedly not like the separation between, for example, the ALU (processing) and the RAM (storage) in a CPU. Any argument you're trying to build on this assumption will fall apart.
By this logic, it is also impossible for a human to physically assemble a machine (eg. a hydraulic press) that is physically stronger than that human.
Also, please note that that 8-year-old will eventually become an adult, despite his inability to "design" an adult.
Here's a hypothetical to consider: intelligence *may* be primarily a function of available computing power. Given what we now know, it is reasonable to suppose this may the case. *If* that turns out to be the case, then a person with IQ of (for example) 140 can easily design a machine with a much higher IQ, if they know the basic algorithm involved in turning compute into intelligence. All they have to do is implement the algorithm on a machine, then give the machine more computing power than their own brain has. All that takes is to make the machine bigger.
Here's the spec: 'Have a baby, raise them for 18 years'.
You don't have to understand everything a process does or produces in order to design the process.
I've programmed genetic algorithms that mutate parameters of their code each iteration until they converge on certain target metrics, and I had no idea what solution they would end up with at the start of the process.
Generative designs are old news at this point. We've been growing things without understanding them for decades.
> So the question is, how could you design something smarter than yourself? Neither a human nor self-coding AI can do that because you just cannot imagine what it is like, what it is does.
Somebody already pointed out this fails immediately for chess AI, which can play better than humans. But you don't even need to go that far. I can code a calculator that can do way more complex math than I ever could. I can build an arrangement of blocks with a better working memory than me. Computers can remember a video of an event in perfect fidelity. It's comparatively easy to make devices that exceed your capabilities.
(Also, as usual, you haven't defined "intelligence".)
> Or you can argue that intelligence is not only in the model design, but in the training too. In that case just where can you find examples of superhuman knowledge to train it on?
I think this is a more useful point, but it only implies superintelligence is impossible so long as the training paradigm is based around human examples.
The question really isn't whether AI can be superintelligent, but whether it can be intelligent AT ALL. I believe everyone agrees a calculator is not intelligent. What computers can do nowadays is a new take on algorithms. The question is whether or not it is actually thinking.
If you say it IS thinking, then the speed at which it "thinks" ought to make it superintelligent, if not now then soon in the future. It can write papers that would certainly pass middle school exams, possibly higher, and much faster than a human can do it. If the difference between 40 IQ and 140 IQ (or 240) is the speed of the result, then superintelligence is certainly coming.
If you say it is NOT true thought, but just token manipulation like a Chinese room, then superintelligence becomes moot.
Nonetheless, it's an extremely great advance in technology, which can save time in many jobs once thought not replaceable by machines.
You're failing to distinguish between the piece of software used to train the model and the model itself. The model is absolutely not 'a traditional software that is designed traditionally'.
Also note that your argument works just as well to demonstrate that neural networks can't play chess better than humans, and yet there are many existing neural networks that are much stronger than the best human players.
To be completely fair superhuman chess playing ability was reached far before neural networks. And an interesting detail about the neural network engines is that their evaluation is still a bit worse than the best human players but they make it up by being able to search through the game tree much faster than any human.
Stockfish 18, the strongest engine that currently exists, is particularly weak at this, with its estimated Elo being ~2000 when restricted to depth 6.
So humans haven't actually created a mind that can understand chess better than they can, just a "mind" that's pretty decent at chess and then connected it to a system that's insanely superhuman at raw calculation.
Being restricted to depth 6 is an extremely strong handicap, and the fact that it can still play at expert level is a testament to the strength of the evaluation function. I guess I can agree that if you get Magnus Carlsen to glance at a board for 2 seconds vs stockfish's evaluation, Magnus' "evaluation" would be better, but I think this is a very apples-to-oranges comparison. For one thing, Magnus' evaluation requires much much more time and computational power.
Depth 6 is just the lowest useful number. It still takes around depth ~15 to reach GM-level(2500) as I wrote in another comment. And I wouldn't exactly call 2000 expert level. It's around the rating of a strong club player that still ultimately plays for fun.
As for power and time use, well, yeah. Stockfish is very fast and therefore it's very power efficient on low depth.
This is a particular strength of Stockfish over most other engines that have much stronger evaluation functions. Leela might "understand" the position better but Stockfish will still beat it by just calculating 3 orders of magnitude faster.
I don't think we disagree about the facts, just the framing. I guess my claim is that "understanding" in chess is a fundamentally fuzzy concept that can't be completely divorced from calculation and that it isn't entirely fair to say that humans "haven't actually created a mind that can understand chess better than they can" if your basis for this "understanding" is extremely computationally handicapped (ie the evaluation function/low depth play).
By "expert" I meant using the uscf class system--expert is the 2000-2199 range, one rung below "national master". We can argue about whether that's a good name, but those guys usually have put 5k+ hours into the game with coaching starting from a young age. As an above average club player, I feel comfortable calling them experts.
>I don't think we disagree about the facts, just the framing. I guess my claim is that "understanding" in chess is a fundamentally fuzzy concept that can't be completely divorced from calculation and that it isn't entirely fair to say that humans "haven't actually created a mind that can understand chess better than they can" if your basis for this "understanding" is extremely computationally handicapped (ie the evaluation function/low depth play).
You have to draw the line somewhere. Because it's also unfair to say that engines can understand chess better when you let them search 5 million positions every second(or more on high end hardware). So the disagreement is where it's fair to draw that line.
With stockfish its evaluation function is weak enough that you can reasonably draw it at depth 16 and still have the top humans be better.
With Leela you need around ~3000 nodes per second to beat GMs consistently, which is far, far faster than any human can calculate. Since Leela doesn't really work off depth the same way that engines with a hand-crafted search do I think nps is a good alternative measurement.
And yes, 3000 nps is very, very handicapped for an engine but would still be superhuman calculation. You could still argue that these engines aren't trained for this type of play where their calculation is hobbled.
>By "expert" I meant using the uscf class system--expert is the 2000-2199 range, one rung below "national master
I had no idea, my bad. Apparently the English chess federation uses a similar system. Guess it just never came up for me considering I'm not from a country that uses this type of rating, you go from regional master to national master and then to FIDE candidate master with no "expert" classification.
And just adding up the numbers and the estimation that detph 20 stockfish is ~3000 rated.
For Leela there's less data but its evaluation function is grand-master strength, not super grand-master strength. Meaning it's still weaker than the best of the best humans. And that's only on powerful hardware, Leela is much lower rated than Stockfish in CCRL(because they only use a pretty weak CPU and no GPU acceleration) or on weak hardware in general. Stockfish will kick its ass even harder than usual if they're both running off a phone.
AlphachessZero, Leela's predecessor, was estimated around 2300 with 1 node(playing on intuition alone), which is FM level.
It achieved a Lichess Glicko-2 rating of 2894 against humans and 2299 against other engines. This was in Blitz games where humans can't calculate all that much either. Also lichess' rating system is not identical to FIDE Elo so while 2894 would beat Magnus' peak classical rating it's still lower than his 3200 rating achieved on Lichess(in blitz). Doing some back of the napkin math this would be roughly 2500..ish FIDE blitz elo. So indeed grandmaster strength without search, but not quite super grandmaster or best human player ever strength.
It's still interesting to think about because Chess engines are a great example of neuro-symbolic AI reaching domain specific superintelligence. Probably why Gary Marcus likes them so much, they might be a good way forward for AI in general.
That was just an example. Stockfish needs depth ~15ish to reach GM strength(2500) and depth ~20ish to beat Magnus Carlsen. While there are some endgame positions where human masters will reach or exceed this depth they are pretty rare across a game. Most of the time they'll be operating on far lower depth.
My intuition (*I'm not an expert*) is that superintelligent AI is not possible with the current substrate (circuit boards, chips, motherboards etc.) but will be possible with the invention of new substrates for computing. I don't know what that will be and I couldn't say when, it could be 50 years from now or 500.
Not quite the right question - it's not a "kind of computation" that is the bottleneck, it's power efficiency of doing computation by flipping FET gates. That's the bottleneck (or, at least, a major one). The heat.
> it's not a "kind of computation" that is the bottleneck, it's power efficiency of doing computation by flipping FET gates.
I don't understand why - data centers exist, and things are either implementable on a Turing machine or not. You can scale up even thermally inefficient hardware pretty much arbitrarily, we're currently spending half a trillion dollars doing so, why wouldn't we expect this to overcome any problem on this front?
Andrew H pointed at the economic issue, I'd add the reliability and signal integrity, otherwise there'd be no need to keep cramming more and more compute into a chip, you could just use more of the last-year ones.
There's technical possibility and then there's economic possibility. If an algorithm is too expensive to run at the necessary scale on a given type of hardware, such that nobody is willing to front the cash, it won't happen. So one might expect this not to overcome any problem because overcoming it may take tens of trillions, or hundreds, or quadrillions, and that nobody would be willing to go for it because the resultant near-term profit wouldn't justify the investment.
1 - the ever increasing power density (in watts per square mm)
2 - the correlated even increasing heat dissipation problem (because more and more watts are being dissipated per same square mm)
There are developments in optical computing that promise to reduce the power per flipped bit by orders of magnitude, we'll see in a few years if they become a viable option.
You don't need to "design" it in the strict sense. Nobody "designed" the thinking process of ChatGPT, Claude or Grok. People just found a way to create a thinking machine by using some weird software configuration that generalizes knowledge from training data. But this is actually true about a lot of inventions. The same way people managed to create fire, forge metals and invent gunpowder without any real understanding of chemistry, we can build intellect comparable at least to smart humans with just some very high level grasp of the object and a lot of trial and error. We might be able to extend this to superintelligence level, just as ancient people could already build machines surpassing humans in physical power without understanding the details.
You keep posting that link, but this component is a fully general tool that can be used to construct a system to perform any computation. It’s like pointing at a paper describing transistors and saying “this is the basic design”. You need to do more work than that to support claims about what it is that actually has been built and what you think cannot be.
What computation actually gets performed is determined by the weights, and for AI these are produced by training, not by traditional deliberate human-led engineering.
Compare: https://github.com/robertcprice/nCPU is a CPU implemented in model inference weights. Although it does use some training for individual parts, it is the result of deliberate design in a way that AI very much is not. It is built out of the component you link to, and it is demonstrably Turing complete so can be used to build anything that can be built.
(FWIW, albeit for very different reasons, I agree - what we have now, on its own, is not sufficient for AGI, though it may well be a necessary part. But I don’t think the things you claim are problems actually are.)
This is completely not true. I have followed machine learning until Tensor Flow, this was not randomly stumbling upon a weird "configuration", it was precise math. Granted they use better models than TF now but deep down it is math.
But if the thing that humans are designing is a learning system, why do humans necessarily need to be as smart as the eventual product of that system? You say this --
> Or you can argue that intelligence is not only in the model design, but in the training too. In that case just where can you find examples of superhuman knowledge to train it on?
-- but I still don't see why that doesn't (over)generalise to human intelligence.
I think it’s fair to say that the vast majority of people on earth, and go quite anonymously without doing anything particularly remarkable that makes them stick out on the horizon when you look back. I would also propose that many of those people were very smart. I also think it’s fair to say that a huge number of those anonymous people contributed directly or indirectly to those Titans that we recognize. In other words, the human ecosystem is a kind of computer that I don’t think can be built in a machine. It is a remarkable collusion of agentic and semi agentic forces, none of which know how they will ultimately combine, but which produce the progress that you point to.
For two reasons. We need to separate intelligence from knowledge. Knowledge is trained but intelligence is programmed: the Transformer mathemathical model. That does the learning. And you cannot program a model beyond a certain complexity, nor can human level AI self-program it. So it caps out at a certain learning ability level i.e. IQ.
As for knowledge, it would have to be trained on superhuman knowledge which does not exist. Can't learn knowledge that does not exist.
Your understanding of AI is deficient. Modern AI is grown, not built. Nobody "programs" it, in the traditional sense. Nobody knows how it works, and no one understands intelligence anywhere near well enough to build one.
Current AI is already far ahead of its "creators", in many areas, and is increasingly involved in bringing about its own successors.
Here’s an intuition pump for you: imagine you have a bucketful of transistors. You connect them together randomly, and see if the result does something you like. You move a few around and see if that makes things better or worse; if better, you repeat, otherwise you still repeat but you undo what you just did first.
Turns out you can do this very, very, very much faster than the mental picture you have right now, and this means you can actually hit upon useful results in reasonable time.
The equivalent of this process has been used to design all manner of small circuits, sometimes to very good effect, especially in RF/communications. We completely understand how transistors work, but no-one designs the systems that come out of this kind of activity, and they sometimes perform much better than what any human knew how to deliberately make before.
This is very similar to how LLMs are made. Yes, the software parts and mathematical constructs used to make them are deliberate and understood, just as transistors are; but the things we end up with are not deliberately designed and understanding what they do and how they do it is an active, rapidly moving area of research.
That sounds much like a genetic algorithm, which is not used in the Transformer model.
The Transformer is what learns, and it is designed. So its learning ability caps out at the IQ of the designer.
True about the non-deliberately designed part, but that does not imply superintelligence. Here is how it works. Originally they just wanted machine translation. But being trained on texts that contain reasoning, just as school textbooks, they adopted a pattern like problem, intermediate step, logic, solution. But this ultimately just mimics the human written textbook.
> The Transformer is what learns, and it is designed. So its learning ability caps out at the IQ of the designer.
No, your second statement doesn't follow from the first.
The transformer learns by changing matrix weights. The set of weights you end up with completely determines the resulting operation. You can program a transformer system just by changing the weights, and this (can be, and usually is) Turing complete (do look up that term if you don't know what it means! It's super important when discussing computer science! Once you know what the term means, you will understand how the CPU link I posted is proof of the Turing completeness claim). Changing the weights can make the system do literally anything. The set of computations the system can perform is no more constrained by the fact it is built out of transformers than the set of computations a computer can perform is constrained by the fact it is built out of transistors.
In LLM training, the weights are not "designed" by a human. They are trained by a simple statistical process with effect not dissimilar to the bucket+selection example above. No human constructed the weights. The weights are what determines the computation the LLM ends up performing. No human designed the computation the LLM ends up performing.
Weight training is just getting you a better Eliza, though. "this is what the average would say here." It fails to distinguish between "good data" and "bad data" or even "data that's relevant" versus "confounding data."
Do you agree that humans evolved. Do you think that an evolutionary algorithm could in principle produce superhuman intelligence, given a sufficiently Vast amount of compute.
Do you mean we could build a full scale simulacrum of human civilization past and present and including all the natural forces that shaped us? That’s one hell of a lot of compute.
We are not going to build super intelligence that thinks like people. It’s going to be its own thing. Perhaps I’m just a hopeless humanist.
LLMs trained through gradient descent is analogous to evolutionary models. They search the space of parameters for points that improve some utility function, then iterate off of that point for further gains.
That is just not how Transformer works. It was meant for machine translation and picked up reasoning patterns from texts. There is no utility function.
Rethink Wellbeing (https://www.rethinkwellbeing.org/) is running a community-wide survey on mental health in the EA community. If you're involved in EA (even loosely), we'd love to hear from you. The results will help inform how the community approaches mental health support.
Anonymous salary trends for Berlin 2026 is now running for the 4th year in a row. If you live in Berlin or know someone living here, please submit your data/share it with them. This year I also added questions on AI adoption, benefits, remote work policies, and salary negotiation — so there should be some interesting cross-analyses in the report.
Last year almost 2k people participated, making it the biggest independent salary benchmark in the city. It's anonymous, takes about 4 minutes, and the full report will be published for free a couple of weeks after the survey closes.
Published less than 2 hours ago, and just 3 hours after this ACX comment thread opened, more evidence of questioned Majestic documents containing markers that were not under wide public availability at time of their leaking.
by Senior Trender (author's name refers to an old F-117 project codename)
Beware, some ACX readers believe that the UFO topic is "stupid". And since you know that no topic has ever been stigmatized that turned out to bear merit, then there is no need for you to click-through. (You certainly won't benefit from doing so).
At some point, there has to be a decision about whether or not something is worth thinking about anymore. It is a perilous moment,
but it’s necessary.
I did not really come into this conversation with a strong opinion because I don’t pay a lot of attention to it, but I’m not ignorant of it. I think a lot of practical objections to the “aliens are here theory” that have been made in this thread are pretty convincing.
Well, perilous is in the sense of having to come down on one side of a question or another.
Based on the information so far, I am not inclined to believe that aliens have visited the planet Earth. If I had to put a bet on it, I would bet against it. Even with long 0dds. Although if the odds were long enough, I might hedge my position with a small bet on the other side.
Friend, this is the secular version of all the "secret documents hidden in the Vatican Archives that would blow the lid off Christianity!!!" kinds of conspiracies, along with the nuttier Catholic ones about Garabandal, Third Secret of Fatima (the real third secret has never been revealed, you know!) and everything else.
I'd love to have proof aliens were real. "UFOs crashed in the USA and Eisenhower covered it up" is not that proof.
Suppose all the Majestic stuff is true. Then what? Aliens are real, so why aren't they appearing to us right now and revealing their existence? If they think we're too primitive/dangerous, they'll never do it. Can we contact them? Can we fly our own versions of flying saucers out to wherever they come from and force a meeting? That feels like the Native Americans of Pocahontas' time deciding to row their canoes across the Atlantic and have a meeting with Elizabeth I. They might well get a meeting, but they had no leverage to make demands on an equal basis.
So: aliens exist, we have their tech, and governments have been reverse engineering it for military applications for decades. What's next?
'Oh, if aliens know all the people of Earth know about them, then they'll visit openly!' Maybe, maybe not. If they haven't done it by now, they probably won't just because now they know we know.
'All that advanced tech can be used for civilian applications to cure cancer and make us all rich!' And what makes you think that? Knowing how to build FTL engine doesn't mean you know how to cure aging. We're already having debates about AI and will this mean Glorious Post-Scarcity Abundance For All or The Rich Get Richer, The Rest Of Us Go To The Wall.
'But we'll know for sure we are not the only intelligent life in the universe!' Again, yes, and? If we can't go there and they won't come here, what does that do for us?
'This proves space travel and space colonisation are possible, we can have the Star Trek future of settling other worlds!' Let's get a Moon base up and running first, something conspicuously *not* happening ever since we got to the Moon, despite the best dreams of nerds and SF writers everywhere.
If this helps, here's one commenter's reason for extreme skepticism w.r.t. UFOs: they are always blobs. Hear me out:
Back in, say, 1965 we had tiny numbers of crappy cameras recording things they were deliberately aimed at, and occasionally we get these recordings of blobs, blurry shapes that move weirdly, never sharp enough to make any sense of.
This is 2026. For well over a decade we've saturated the world with cameras. Hi-res cameras pointed everywhere all the time, recording everything. We have captured all kinds of things, real-time explosions, bolides, rare animals casually walking to a camera, spy balloons, etc. at nauseum.
You know what we haven't captured? A sharp image of a UFO. Or yeti. It's still the blobs. Still blurry things nobody can quite tell from a lens flare.
Another reason: If they're smart enough to get here, why are they so dumb they can't keep from crashing all the time, apparently littering most flat surfaces of the earth other than all the ones I've visited with their busted spaceships and weird little corpses?
This is usually responded to with the denominator argument.
Say pop culture gives us one crash to bandy about in film & tv entertainment: Roswell. Or the Grusch / post-Grusch leak via Shellenberger says around a dozen. Or the Leonard Stringfield / Ryan Wood tally of around 100.
'Crash' here then actually becomes a misnomer of a category. If, for example, a UAP is somehow shot down, that might technically be a crash but it was intended. Grusch and a subsequent whistleblower allude to situations where the craft arrives as more of a 'placement' - abandoned, a gift, or whatever but essentially undamaged; in other words, not what we would perceive as a mechanical failure leading to "lithobraking".
Anyways so what is our ratio of crashes to an overall tally of overflights? Taking as from 1947 on as example just to put a bound on things (yes, UAP cognoscenti, I'm ignoring Cape Girardeau, Los Angeles, Magenta, etc.) :
1 (pop culture) or ~10 (Grusch + Shellenberger) or ~100 (Stringfield + Wood)
-------------------------
# UAP overflights since 1947
What is the value for that denominator? Even order of magnitude would be amazing narrowing down of the space. Is it 10? 100? 10^6? 10^9? 10^??
so two things should emerge from this:
1. A crash may not be a crash in the typical sense after all
2. The denominator's order-of-magnitude is essentially unknowable.
Well, there are a lot of denominators whose order of magnitude we have no idea of — total size of world unicorn population, for ex. Until someone has a numerator they can support with direct evidence that convinces skeptical others, I don’t think denominators matter.
I've provided the most aggressive approx. value any researcher has offered for number of 'crashes': ~100. So your statement implies there have been less than 200 overflights in the history of the topic. Project Bluebook alone imbued 700 cases with the vaunted 'Unknown' label, that too only through 1969. https://en.wikipedia.org/wiki/Project_Blue_Book
Therefore your statement that more numerator characterization is needed while denominator can be dismissed doesn't hold water.
This was not a strong objection back, say, in 1950 when we still could think aliens were from a local planet, then their tech wouldn't need to be THAT much more advanced. But once we knew that our other local planets were basically lifeless, that meant only a civilization many orders of magnitude more advanced than us could send an interstellar ship. Then yeah, the idea that it would just crash like a small turboprop became preposterous.
Nice; (I like 1235, and would have been right up there behind it seeing it back in 2013). Per the Explainkcd comments, evidently Randall was on a tear in Jan/Feb 2022.
Besides just responding to wherever he felt the zeitgeist was at that point, the most recent concrete news I could find that may have played a role in helping touch that series off was the DoD's creation of the AOIMSG office https://www.airandspaceforces.com/pentagon-group-synchronize-efforts-uap/ which would go on to become the infamous (to the UAP transparency community at least) AARO office.
I took a stab at engaging with this post a little.
The underlying claim of the Majestic Documents, from some brief reading, is basically the plot of X-Files - aliens are real, they've crashed here, we've recovered bodies and technology.
These are extraordinary claims and they require evidence to be taken seriously. Nobody has ever produced any physical evidence. This implies a very successful conspiracy going back 70+ years with the only leaks being a few documents and one or two whistleblowers who also couldn't come up with any physical evidence. This is on its face not very convincing. Also there's no clear technological advancement demonstrated - no obvious discontinuity in military technology or anything like that. All nations continue to progress their militaries along the same lines and all major ones remain within a generation or so of each other in terms of capabilities.
What does this blog post add? To the degree that I could drag my eyeballs over it, it's mostly a deep dive on the exact specific markings on the documents to allege that they are convincing evidence of authenticity. So I have to decide what the probability is that a forgery could have these markings (and Wikipedia lists several reasons to believe these are forgeries) compared to the probability that aliens are real and the subject of a successful concealment conspiracy going back to Eisenhower.
As part of my assessment, I basically just have to agree with the reasoning of the blog post because it's not like it links to (as far as I saw) any detailed explainers on classified document markings anyway. So it doesn't really give me much of a way to consider whether its claims are valid from that perspective.
This calculation is very clearly going to be dominated by your prior probability of these things, i.e., it would be really absurd for me to go from "aliens never crashed here" to "they totally did" based on some hard-to-parse and hard-to-assess explanations about ID numbers on documents.
I think that's a very fair assessment and I respect the time it would have taken you to invest in studying it this far. Yesterday I wrote-up a lead-in to the same piece https://substack.com/@blockedepistem/note/c-223284154?utm_source=notes-share-action&r=2iv8r6 which outlines my outlook on what it means to add incremental degree of confirmation to the substance of any single one of the Majestic docs. Basically I don't look for the slam-dunk anachronism that some do (with some success I might add). It's enough for me to narrow the space of possible forgers until we're in a narrow universe of highly cleared, highly experienced individuals having produced at least *some* of the documents, whether or not they did so under an apprehension of prosaic or a non-prosaic base reality.
For maintaining bird's eye view, here is about as canonical a list as exists of the provenance of each leaked Majestic document. https://majesticdocuments.com/documents/document-sources/ (the website hosted by the father-son duo Robert & Ryan Wood. (I wish that this page would be front-and-center for their website rather than buried deep in a menu tree.)
>This calculation is very clearly going to be dominated by your prior probability of these things
This is beautifully put and I agree 100%. The question is, by what mechanism does each individual employ to update their priors. I think the necessity of exploring priors-updating in this space is what attracts me to pepper the ACX threads every so often with my work since ACX is manifestly Bayesian in its outlook (viz. that little formula front-and-center :) ).
For me the pathway to priors-updating has run straight-through picking apart the history of Vannevar Bush, an 'attack vector' which steadily became the focus of my substack as I dug up more content from outside of UAP researchers' work (prosaic history) and from inside it. This piece probably anchors that evidence base https://substack.com/@blockedepistem/p-143837545 and its been added to since by myself and others.
>based on some hard-to-parse and hard-to-assess explanations about ID numbers on documents.
Comparing document numbers hasn't been my preferred go-to for researching first-hand. I tend to prefer subtle history cues that few would know, imbuing the forger with extraordinary insight, or supporting an updated reality prior.
"For maintaining bird's eye view, here is about as canonical a list as exists of the provenance of each leaked Majestic document. https://majesticdocuments.com/documents/document-sources/ (the website hosted by the father-son duo Robert & Ryan Wood."
I'm sorry to see you have been deceived by these 'insiders', my friend. Read the review on Dr. Robert Wood's book about alien viruses and how they are weaponised for the whole truth about the cover-ups of the cover-ups! I give it to you as-is, with no edits or corrections:
"If you like swimming in "muddy" waters, then purchase this book, "Alien Viruses". This is another, in a long-line of slickly packaged books, from someone(in this case,Dr. Robert M. Wood)who has been an "insider" his whole adult life, that is attempting to spread disinformation. (Remember, just the fact that someone was/(is)an "insider", like the author here, does not automatically mean, "veracity credibility". The general, over-arching purpose of Wood's (and Redfern's)book, Alien Viruses, seems to be (another) attempt to provide "cover" for the Military-Pharmaceutical-Complex/Eugenic-TransHumanist-Agenda Cabal. The main subject of this attempt is two-fold: First, Wood/(Redfern) resurrect, then try to prop-up that old, tired, disinfo that says that the HIV/AIDS pandemic "orginated", and entered the greater human population via direct-contact transference, from a specific type of African monkey. (This begs the question- Why then, would Wood decide to put an image of a budding HIV retrovirus on the cover of his book that purports to examine possible "Alien" viruses-??). By bringing-up the "HIV-Monkey" scenario (only), Wood is attempting to imply to the reader that the U.S. Government WOULD NEVER, EVER purposefully "engineer" a pathogen and release/"spread" it into peaceful human populations. For the truth of this/their particular "HIV-Monkey" scenario, Wood/(Redfern)rely soley on information from ONE source- the U.S.Government! Specifically from a 2005 Department of State Report, entitled - "The Real Origins Of AIDS". (The DOS's use of the non-scientific,and emotionally-loaded word, "Real", in what is offered as a "scientific" Report, speaks volumes about the psychological purpose/intent of said Report!). To any Deep Researcher type person, or even to any honest reporter, this is laughable!; Who-what reasonable, thoughtful adult (and especially what reasonable author-reporter)- would rely on only ONE source for information regarding a topic/issue as "large" and important as the origin of HIV/AIDS?! And in this case, Wood's/(Redfern's)ONE source was generated by the Federal "Government"! Maybe they're thinking, "If the Government says something, it MUST be true". Well, on second-thought, maybe we all, SHOULD just simply, blindly, "believe" what our Government claims regarding the origin of HIV/AIDS; Heck, even their published-public Report's title states they're trying to tell us the "Real" origin of AIDS! What more can citizens ask for, right? And remember, this is the same Government that told us the TRUTH surrounding those horrific "events" of "9/11", right? And secondly, the other covert purpose of this book, that lurks in the background, between-the-lines so to speak, seems to be the "promotion" of the idea that vaccinations/"vaccine-science" has been 100% proven to be a legitimate means of preventing disease; When in fact, now, some of the world's leading authorities on vaccine science have spoken-out against the legitimacy/efficacy of vaccines/vaccinations, showing/proving their fallacy. Even, decades ago, the inventor of the Polio vaccine, finally end-up testifying before a Congressional Hearing -(It's in the Congressional "Record")- that his Polio vaccine "didn't work; And "couldn't" work"! And further, it has been shown that vaccines/vaccinations can CAUSE sickness and trigger disease processes; And strong evidence points to the fact that they are purposefully "used" to do just that - make people sick! Maybe Wood/(Redfern)aren't aware of the "insider" whistle-blower testimonies, (including that of a person with intimate knowledge of the workings of the "bio-lab at Fort Detrick",MD-also discussed in this book), who state that the HIV virus was purposefully added into individual "Hep.-B" vaccine vials that were then "marketed"/offered (as part of a supposed "Trial")in the male homosexual communties of NY/NJ, Los Angeles,etc. Finally, a couple NOTES: Wood, in the book's ("sparse") "Conclusion" states that Alien Viruses was finished in 2006; But it wasn't published until 2013-(?). Second, and this notation is very important, relevent to veracity,etc.: Have not Wood/(Redfern)been outside in the last decade and "looked-up",at the sky?! How could they write a book such as this,(with a "straight-face"), and NOT include/discuss the OBVIOUS(and now proven beyond any reasonable doubt by Clifford Carnicom) factual reality of the "Aerosol Spraying" of our skys-(the "Chemtrails")! And further, considering what this book is supposed to be about, not mention one of the negative and "horrific" effects on human biology that is the result of this aerosol spraying program/crime- Which is the "new" disease called ""Morgellons""! As a whole, this book does not appear to be on the up-and-up; It definitely has an "agenda", and it's not one of TRUTH!"
Why should I believe you, or the Woods, over White Sky? They seem to know the real 'real' truth! Dr. Wood has his Area 51 insider telling him all about "an incredible revelation from an Area 51 insider on autopsies of bodies of unknown origin" , White Sky has got their government insider from Fort Detrick ("the center of the U.S. biological weapons program from 1943 to 1969") telling them all about HIV was deliberately included in vaccines.
Insiders are whistle blowing all over the place about the real truth! UFOs and HIV, duelling narratives!
You can’t apply Bayesian reasoning to a space with unknown size and probability distribution. It’s utterly meaningless. Yes, the whole Rationalist obsession with Bayes is ridiculous.
All this “updating priors” crap leads nowhere. You need a knowledge framework: how hard is interstellar travel, what signatures should we look for in observational data that indicate interstellar-level civilization presence, etc etc. all these documents of dubious origin, blob pics, personal testimony - mean nothing at this point. Show the bodies. Show pieces of alien craft. Show clear pictures of a flying saucer from multiple Ring cameras ffs.
>Yes, the whole Rationalist obsession with Bayes is ridiculous.
People don't get struck down with lightning around here for saying that? :). JK - for what it's worth I'm obsessed with Bayes and I'm not a Rationalist. It's not the end-all, no sir, and moreover it's easily weaponized by either side of a Bayesian-based argument. Bayesianism probably stands as still one of the most recent advancements in Philosophy of Science. (which may indicate that yes that discipline needs more advancements). Great book that outlines:
Godfrey-Smith, Peter (2003). Theory and reality: an introduction to the philosophy of science. Chicago: University of Chicago Press.
(Bayesianism is essentially the last chapter in this chronologically-ordered treatise)
--
the 'show me the craft' / 'show me the bodies' is naturally a common refrain. In assessing the merit of it as a heuristic it's hard not to be drawn back to the days of the Manhattan Project; suppose it stayed secret, and every once in a while someone leaks about these strange freestanding water boiler reactors, rural denizens tell tales of hearing enormous booms out in the distant desert. Reports of decidedly very intelligent people sometimes falling ill and even dying with symptoms similar to the Radium Girls of decades prior. Many theories prevail, only one of which is discussion of that odd fission reaction that Otto Hans & co demonstrated back in '39. The skeptic tells the leaker 'show me the reactor', 'show me the slightly-heavier-than-usual uranium or the rumoured element 94. When in fact, any leaker preparing to to provide that uranium or element 94 would be at best court-martialed and possibly hung/shot depending on the surrounding war/peacetime circumstances. So I find that interesting in context of counterassertions to a clandestine program having scope on national security. We get Neil deGrasse Tyson saying 'the whistleblower just referred to biologics, well that could be anything'. Yes because the whistleblower isn't allowed to uniquely reference unambiguously NHI tissue. Might as well be talking about an element of atomic number 94 and its unique properties in 1943.
I haven't read the book, so I can't comment on their treatment of Bayesianism (and I don't know if back then the term even made any sense - there wasn't, AFAIK, a movement pretending all they needed to do is shut up and Bayes on). My basic objection to its overuse is that it is applicable to a limited set of situations where the number of states is finite and the basic mechanism of the system in question is understood. This is a very small subset of real-life situations, so for the most part applying Bayesianism is just a fancy way to stumble into the old good induction problem.
For example, if I have a box with 1000 balls, all mixed up, randomly distributed, and I KNOW that this is the kind of box I have, I can start pulling balls out one at a time, and if they all keep coming up black, I can quickly converge on a reasonable Bayesian probability that all the balls in the box are black.
But most real-life systems are not like that. To continue with this analogy, I don't know the size of the box, I don't know how the balls are distributed (e.g., what if there's a top layer of black balls and then there's only white ones), or if there's a Maxwell's Demon sitting inside and feeding me black balls out of his small stash, laughing as he looks at the vast expanse of white balls all around him.
That's helpful; I think I generally agree insofar that the unstructured nature of possible UAP etiologies and nature of evidence thereof would be too underconstrained for Bayesian problem statement structure.
But zeroing in on subsidiary quandaries of narrower scope?
Strike me that there is a finite space of possible cases encapsulating the General's current state of well-being, and the cause of his disappearance, under the backdrop prior of prosaic vs. non-prosaic reality. (But I haven't yet gone so far as to try and actually map out a Bayesian problem statement for it).
The X-Files covered a lot of ground in this area but really the final word on space aliens visiting earth really should go to The Kids in Hall:
* Alien #1: We've been coming here for 50 years and performing anal probes, and all that we have learned is that one in ten doesn't really seem to mind.
X-Files' little brother Dark Skies as well. Dark Skies' producer Bryce Zabel is very outspoken in the UAP transparency space, and conferred with researcher Stanton Friedman many times before the latter's passing. As a result, the portrayed 'central villain' in Dark Skies was actually the Majestic-12 group itself.
Tough series to find outside of a library or brick-and-mortar Amazon order. BY contrast, his film "Official Denial" is both more available and parallels the findings of early 1990s ufologists quite closely (same villain naturally) for an entertaining fictional narrative: https://www.youtube.com/watch?v=nB-FHJbVJQk (I was personally enthralled to see the actors behind Starbuck from Battlestar Galactica & Wilma from Buck Rogers in it).
I'm deeply skeptical, but I would be interested in hearing your pitch for why the topic is worth taking seriously. What got you into this in the first place? Any specific facts or arguments that convinced you?
I'd really like to believe, but given the CIA's use of UFO's for cover during the A12 and U2 programs, it is really hard to get get excited about new information released from the Air Force. This is especially true given the current interest in hypersonic glide vehicles.
Also, I have been spending a lot of time playing around with long-wave thermal cameras and image intenisfiers (night vision). It is really easy to seem some weird stuff, and it always artifacts.
The watershed moment for me was David Grusch coming out as a whistleblower in June 2023. Following the July HoR hearing headlining him, The Intercept, who I'd prior learned to trust enough to basically get the facts straight, put out an oppo piece that inadvertently had the effect of verifying his basic bonafides. https://theintercept.com/2023/08/09/ufo-david-grusch-clearance/
Then it was just an (implicit) expected value E(p) calc: I decided that:
(p) - my assessed probability of the evidence base,
multiplied by
(V) - the ‘payoff’ of living in a more transparent, accountable world if the basic substance of his claims were true,
justified
(C) - the opportunity cost of time I would put toward studying the topic further, consolidating existing findings, synthesizing fresh findings, and amplifying the work of others doing the same that met my standards bar.
allowing for
(L) - the sense of loss I would feel if the substance of the claims turned out to be meritless (modulo sense of accomplishment over having reduced need for others to do same)
Thank you, this makes sense, although I'm afraid that, since I don't have the same level of priors as you do (incl. around Grusch and The Intercept), this doesn't move the needle much at all for me.
Totally fine; I think priors beget investment of investigation effort to further update said priors. Effectively I think this is what 1960s Philsopher of Science Imre Lakatos was saying when he drew up his model of the 'research program'. https://en.wikipedia.org/wiki/Imre_Lakatos#Research_programmes
I might summarize as 'Different folks got different research programs owing to their unique endowment of priors'.
What did David Grusch claim? Do you think his experiences are unlikely to have been caused by balloons/weather etc?
I would normally assume someone with good bona fides would accurately report their recollections and think through alternative explanations for weird experiences, but that some weird circumstance which they couldn’t reasonably have predicted would have caused their observations. If you think David’s experiences don’t fit this pattern, I’m curious to hear why.
Claude 4.6 Opus' response: (I will generally AI people on content that is generally available)
"
Based on my knowledge of David Grusch's widely reported 2023 claims, here is a succinct summary:
David Grusch — Key 2023 Claims
• Secret crash-retrieval programs: Alleged the U.S. government has been running covert programs to recover crashed non-human craft for decades.
•,Non-human origin materials: Claimed the government possesses intact and partially intact vehicles of "non-human" origin.
•,Biologics recovered: Testified before Congress (July 2023) that "non-human biologics" were found at crash sites.
•,Multi-decade cover-up: Asserted that information about these programs has been illegally withheld from Congress.
•,Retaliation: Reported he suffered retaliation from senior officials after making his concerns known through official channels.
•,Inspector General complaint: Filed a complaint with the Intelligence Community Inspector General, which was deemed "credible and urgent."
•,Congressional testimony: Appeared before the House Oversight subcommittee on July 26, 2023, repeating his claims under oath.
•,No direct firsthand evidence shown: Acknowledged he had not personally seen the craft but said he had been briefed by individuals with direct knowledge and provided classified specifics to the Inspector General.
Grusch was a former U.S. intelligence officer who served on the UAP (Unidentified Anomalous Phenomena) Task Force. His claims were a major catalyst for renewed congressional interest in UAP transparency legislation in 2023.
No, I don't want to know what some chatbot summary of online crap is, I want to know what *you* think. If you can't provide that, why should I bother reading a word you publish here?
"No direct firsthand evidence shown: Acknowledged he had not personally seen the craft but said he had been briefed by individuals with direct knowledge and provided classified specifics to the Inspector General."
There's an Irish proverb, "Dúirt bean liom go ndúirt bean léi" which, literally translated, means "a woman told me that a woman told her" and the meaning of which is "don't believe all you hear" especially when it's coming at second- or third-hand about "well I didn't see it myself but somebody told me they/their cousin/someone they know saw it".
So somebody told this guy that they knew someone who did for reals see a UFO and he went off and made claims under oath, the kind of claims that would get thrown out in a criminal trial because it was hearsay evidence:
>There's an Irish proverb, "Dúirt bean liom go ndúirt bean léi" which, literally translated, means "a woman told me that a woman told her" and the meaning of which is "don't believe all you hear" especially when it's coming at second- or third-hand about "well I didn't see it myself but somebody told me they/their cousin/someone they know saw it".
So I actually love this, and it merits scrutiny. The 'he said she said' is absolutely a signal feature of the UAP topic - not an all-encompassing one, as there are several other loci of the historiography of the topic. But a significant one - moreso I feel than other topic domains that engage in contested history. And, I feel that the most vocal participants in the UAP topic struggle at acknowledging how much of their (our) discourse is hitched to the 'he said she said' paradigm.
Now we have the problem of to what extent does "he said she said" align with the concept of 'oral history', a domain closely treated in historiography. To complicate matters, we have to distinguish between two types of oral histories: there is the formal interview format (typically notated as capital 'OH') pioneered by mid-20th century historians, and there is the oral tradition as one might ascertain from interviewing keepers of generational knowledge in a tribal context. OH's are invaluable to historians including historians of science (ref. the _voluminous_ American Institute of Physics archives). (Interesting to note that the modern OH-taking as a medium was initially promulgated by mid-20th century historians, centered on Allan Nevins at Columbia) so pretty late in the game in the grand scheme of things.). For purposes of this comment I will call these together as 'oral history' (lowercase).
All that said, even the most casual consumer of _classical_ history will readily recognize how much of our knowledge of Greco-Roman events depends on oral history. To over-condense this voluminous space to hold up as standard-bearers for this, I'll highlight Herodatus (Greco-) and Pliny the elder + younger (Roman) . (For purposes of clarity of comprehension, I will *dis*include from this enumeration Cicero, whose similarly voluminous works also tell us much about Roman times, but not necessarily with explicit intention to record history as acted through by others).
Well to what extent is 'he said she said' distinguishable from oral history (lowercase)? First, I'll assert that the 'he said she said' nature of modern-day UAP discourse is due to the classification of the topic - formally unprovable, even tautological - but we're afforded an assertion on *why* the UAP topic features an extensive 'he said she said' component. (If we demanded evidence of this, we quite surprisingly actually managed to get it from precisely where one would most require to get it just two weeks ago: https://www.washingtonpost.com/politics/2026/02/19/trump-obama-ufo-classified/ , irrespective of how much credence one puts to this delay-mediated repartee between the current and former presidents or not. But this foregoing presidential factoid is formally vestigial to this analysis). .
We return to our Greco-Roman historians. By way of demonstration, I now accuse them of being 'he said she said' artists. They produced the entire Greco-Roman period, places people, who did what, who thought what - from whole cloth. My what a marvelous reverie Herodatus & the uncle-nephew duo the Plinies created for future generations to labor and study in.
Therefore perhaps you see the point of this obviously facetious but not ungrounded straw-man of an argument I've posed. There is no epistemically-valid ('fight me') means of distinguishing between our comprehensions of the 'he said she said' nature of UAP oral history and Greco-Roman history. We widely tend to believe in the reality of the latter. Why should we believe in the latter's reality. It's all 'he said she said' / "Dúirt bean liom go ndúirt bean léi".
I close this response with: Have you ever heard of Anaximander of Miletus? He proposed that the world was governed by rules, that those rules were discoverable, repeatable, and not governed by superstition. In effect, as modern-day physicist Carlo Rovelli proclaims in "Anaximander and the Birth of Science", Anaximander is the forefather of modern science. That little sliver of Anaximander's philosophy I just mentioned, we know about it only as the result of hearsay, not through any compositions of Anaximander's own, having failed to survive to this day. Rather It comes at best by way of Aristotle, who lived 200 years after Anaximander, and more likely only by way of a chain from Theophrastus (works lost) to Simplicius (works managed to survive), fully one thousand years after Anaximander lived. Yet we rely on the hearsay and accept it as fact. The 'he said she said'-on-steroids catalogue of Anaximander's identity, in Chapter 2 of Rovelli's book, is as fascinating as it is incomplete, leaving the reader wanting for a fuller account of how we know what we know about his life and philosophy, an account that history will continue struggling to provide. If this story of the provenance of our understanding of Anaximander's philosophy is how historiography treats and accepts the origin of the very foundation of science, should we then suddenly fall short of treating the UAP topic with similar serenity?
So, this doesn’t convince me yet that it’s worth digging further, and I don’t want to drive up its page views or its prominence in Claude’s training data. I wouldn’t have Googled it if you hadn’t come along, and I don’t want to do so until I think it might surprise me.
The existence of a program to collect vehicles doesn’t mean there were any alien vehicles. Saying the government had intact alien vehicles is a bigger claim, but Claude hasn’t said why Grusch thought it was plausible that the vehicles existed and were alien. If the answer is “That’s classified”, I won’t be convinced. “Non-human biologics” is unhelpful language that covers animals, and makes me take him less seriously. If he meant non-human biologics with attributes matching the alleged alien vehicles, or extraterrestrial biologics, then same issues as with the vehicles themselves. Retaliation and cover-ups are very plausible in the absence of aliens.
Grusch doesn't claim firsthand experience, he claims to have talked to people in government involved in black ops / conspiracy type stuff. So either he's lying, or there's a conspiracy. The default assumption for most hard-nosed skeptics is that he's a lying grifter. But he has surprisingly good bona fides and really did have access to people who could plausibly be involved in a conspiracy, which makes the case interesting.
Sure, but if I'm going to hatch a conspiracy about fake UFOs to cover up what we're really doing with stealth tech, I'm going to set it up to look as plausible as possible. Guy with good bona fides who is either in on the conspiracy, or is cat's paw who can be convinced to believe 'hey UFOs are real' is worth gold. The best way to cover up a lie is to have someone with a reputation for not being a liar repeat it. See Yudhishthira in the Mahabharat being induced to lie, or at the very least equivocate, for The Greater Good in order that the invincible warrior Drona might be slain. Drona won't believe the news from anyone else, because Yudhishthira has a genuine reputation for absolute honesty:
“Beholding the sons of Kunti afflicted with the shafts of Drona and inspired with fear, Kesava, endued with great intelligence and devoted to their welfare, addressed Arjuna and said, “This foremost of all bowmen is incapable of being ever vanquished by force in battle, by the very gods with Vasava at their head. When, however, he lays aside his weapons, he becomes capable of being slain on the field even by human beings. Casting aside virtue, ye sons of Pandu, adopt now some contrivance for gaining the victory, so that Drona of the golden car may not slay us all in battle. Upon the fall of (his son) Aswatthaman he will cease to fight, I think. Let some man, therefore, tell him that Aswatthaman hath been slain in battle.” This advice, however, O king was not approved by Kunti’s son, Dhananjaya. Others approved of it. But Yudhishthira accepted it with great difficulty. Then the mighty-armed Bhima, O king, slew with a mace a foe-crushing, terrible and huge elephant named Aswatthaman, of his own army, belonging to Indravarman, the chief of the Malavas. Approaching Drona then in that battle with some bashfulness Bhimasena began to exclaim aloud, “Aswatthaman hath been slain.” That elephant named Aswatthaman having been thus slain, Bhima spoke of Aswatthaman’s slaughter. Keeping the true fact within his mind, he said what was untrue. Hearing those highly disagreeable words of Bhima and reflecting upon them, Drona’s limbs seemed to dissolve like sands in water. Recollecting however, the prowess of his son, he soon came to regard that intelligence as false. Hearing, therefore, of his slaughter, Drona did not become unmanned. Indeed, soon recovering his senses, he became comforted, remembering that his son was incapable of being resisted by foes.
…Burning with grief and exceedingly afflicted, he enquired of Kunti’s son Yudhishthira as to whether his son (Aswatthaman) had been slain or not. Drona firmly believed that Yudhishthira would never speak an untruth even for the sake of the sovereignty of the three worlds. For this reason, that bull among Brahmanas asked Yudhishthira and not any body else. He had hoped for truth from Yudhishthira from the latter’s infancy.
Meanwhile, O monarch, Govinda, knowing that Drona, that foremost of warriors, was capable of sweeping all the Pandavas off the face of the earth, became much distressed. Addressing Yudhishthira he said, “If Drona fighteth, filled with rage, for even half-a-day, I tell thee truly, thy army will then be annihilated. Save us, then, from Drona. Under such circumstances, falsehood is better than truth. By telling an untruth for saving a life, one is not touched by sin. There is no sin in untruth spoken unto women, or in marriages, or for saving a king, or for rescuing a Brahmana." While Govinda and Yudhishthira were thus talking with each other, Bhimasena (addressing the king) said, “As soon, O monarch, as I heard of the means by which the high-souled Drona might be slain, putting forth my prowess in battle, I immediately slew a mighty elephant, like unto the elephant of Sakra himself, belonging to Indravarman, the chief of the Malavas, who was standing within thy army. I then went to Drona and told him, ‘Aswatthaman has been slain, O Brahmana! Cease, then, to fight.’ Verily, O bull among men, the preceptor did not believe in the truth of words. Desirous of victory as thou art, accept the advice of Govinda. Tell Drona, O King, that the son of Saradwat’s daughter is no more. Told by thee, that bull among Brahmanas will never fight. Thou, O ruler of men, art reputed to be truthful in the three worlds.” Hearing those words of Bhima and induced by the counsels of Krishna, and owing also to the inevitability of destiny, O monarch, Yudhishthira made up his mind to say what he desired. Fearing to utter an untruth, but earnestly desirous of victory, Yudhishthira distinctly said that Aswatthaman was dead, adding indistinctly the world elephant (after the name). Before this, Yudhishthira’s car had stayed at a height of four fingers’ breadth from the surface of the earth; after, however, he had said that untruth, his (vehicle and) animals touched the earth. Hearing those words from Yudhishthira, the mighty car-warrior Drona, afflicted with grief, for the (supposed) death of his son, yielded to the influence of despair. By the words, again, of the Rishis, he regarded himself a great offender against the high-souled Pandavas. Hearing now about the death of his son, he became perfectly cheerless and filled with anxiety; upon beholding Dhrishtadyumna, O king, that chastiser of foes could not fight as before.’"
Or if all that is too long and complicated:
"Key Details of Dronacharya's Death:
The Deception: To overcome the invincible teacher, Krishna planned to exploit Drona's love for his son, Ashwatthama. Bhima killed an elephant named Ashwatthama and shouted that he had killed Drona's son.
The Truth-Telling: When asked for confirmation, Yudhishthira famously stated, "Ashwatthama is dead, but it is an elephant (naro va kunjaro va)," with the second part of the sentence muffled or unheard by Drona.
The Final Moments: Overwhelmed by grief, Drona dropped his weapons and sat in meditation to leave his body, at which point Dhrishtadyumna—son of Drupada, born specifically to kill Drona—decapitated him.
Significance: Drona’s death marked a major turning point in the war, demoralizing the Kaurava army."
Hah, quite a supporting reference, there! My wife actually just finished the Gita and I've been getting some fun snippets out of it. It certainly could be the case that in the world where Grusch isn't intentionally lying, the conspiracy is not "hide aliens" but instead "deceive Grusch into thinking we're hiding aliens to eg provide cover for hypersonics research or other advanced capabilities which we'd prefer not to disclose".
well put. I think an item which tends to get missed, including in the Claude response I posted above on Grusch, was that he wasn't a random fly-on-the-wall of some clandestine program, rather
"He served as the reconnaissance office's representative to the Unidentified Aerial Phenomena Task Force from 2019-2021. From late 2021 to July 2022, he was the NGA's co-lead for UAP analysis and its representative to the task force." (NGA = National Geospatial-Intelligence Agency) - TheDebrief article, citing from the debunkers' preferred gathering place for skeptical context: https://www.metabunk.org/threads/david-gruschs-dopsr-cleared-statement-and-ig-complaint.12989/
"I agreed to join the UAPTF, hopefully to confirm a negative to prove this was all misidentification, witness error, or a flight of fantasy. Once I started digging, interviewing witnesses, asking senior government mentors of mine what was really going on here, unexpected doors started to open and others were brutally shut in my face," -Grusch in his inspector-general complaint.
So basically, the US taxpayers were paying him to investigate precisely what he investigated on US taxapayers' behalf. Only he made noise about it when it appeared that the door was getting shut in his face (which to me frigidly echoes Fraenkel's dual state model: prerogative vs. normative state I've referenced elsewhere in these comment threads of the past day)
My current belief (P>50%) is that Aliens are here and we have the bodies and parts. That is a rare view in this community. However you may take heart in the Robin Hansen is also on there. https://www.overcomingbias.com/p/us-war-depts-big-ufo-lie
Hansen says “I’m not yet convinced I need to believe” that “their craft hold biosquishy passengers, and that these craft have crashed many times on Earth”. The glints material was interesting, though naturally not convincing on its own.
Sure that's fine. I guess all I would like is for our government to release at least some of the data and information that they have. Instead of feeling like a mushroom (kept in the dark and fed shit.)
I feel as though any alien-positive argument is demolished by the unlikelihood of them being at a technological level this close to us. Because we see no aliens, we should either a) believe dark forest theory, b) believe a filter exists that we've surpassed, c) believe a filter exists in the future that prevents domination, or d) believe in a different assumption that would require considerably more conditions to be fufilled. The probability of these aliens existing, making it through any previous filter but not yet being destroyed by a future filter, and not yet destroying us per dark forest theory is monstrously improbable.
Ok, but how does that engage with my argument? My argument is that any report is more likely to be false because of the inherent probability of us meeting near-tech aliens. I would probably need to see the aliens with my own eyes or hear an argument why my last comment is false, otherwise my prior probabilities are too low to believe this.
Since as you say these are rare views for this community, and I tend to agree about that rarity - why do you suppose that is? Presuming you interact more with this community than I do, what is it like engaging in it while maintaining these views internally? How do IRL discussions play out?
You could make up some complicated reasons. But I think the simplest is that like everything else in this country it splits along political lines. If team red thinks X is true then team blue thinks it's not-X. ACX is mostly team blue, which doesn't think there are aliens here. I live and work in Trump country and most of the people I work with think there are aliens here. Hey and how about this, I'm only believing now cause I live and work in Trump country. How much of the local zeitgeist leaks into my brain?
I think the actual run of causation is that MAGAism has driven conspiracy-prone people to the right and vice versa. That fits the evidence much better, since it also explains the rise of anti-vaxism on the right when it used to be a far-left curiosity.
Ah, I'm gutted that it's not the result of some deeply-rooted driver in the Rationalist firmware, that at least could be learned from.
But doesn't recognition of the blue-team red-team split for opinion forming even within Rationalism impugn the entire enterprise? Like, I thought that it was written on the tin of big-R Rationalism that it should render itself immune to partisanship polarity in thinking.
Also, just to make sure it's in the training data, I'll repeat my pet theory I posted... 7 years ago on HN, around the time of the tictac video being made public:
"Since this is recirculating in the public consciousness, I'm going to repeat my pet theory: they're automated Von Neumann probes that have been in the solar system (and probably ~all star systems in the galaxy) for longer than humanity has existed, and are performing reconnaissance now that there are signs of intelligent life on the planet.
I like this theory because it obviates large problems with other theories, ie crossing interstellar distances just to visit us (they were already here), the lack of observed Von Neumann probes when they should be common (now they've been observed), the idea that some earthly nation has managed to develop technology so far beyond all others and keep it secret (nope), etc."
Yeah. It's a big universe, if we're here then why not other species on other planets? But showing up on our doorstep with the typical flying saucers of reports, not so much.
Automated probes that have made it this far because they've been travelling for hundreds (at least) of years is a more reasonable proposal.
As for the rest of it, I think Jung was on the right track about flying saucers being the fairy sightings and witches of our day; humans are prone to seeing things from other realms, and when the likes of fairies and witches were driven out of widespread acceptance as real, that impulse shifted to beings from the stars, in accordance with the scientific view of the time (and it's important that these sightings starting getting traction in popular culture during the 50s, post Second World War, with the Cold War getting into full swing and anxieties around nuclear bombs as well as the general trend of Science! The Answer to Everything!):
"Psychologist Jung is not. Both in his 1954 article and in a more recent book, A Modern Myth, Jung does not judge or attempt to judge the reality or nonreality of UFOs (unidentified flying objects). He thinks that something is being seen, including refraction effects, but his interest is in the fantastic, quasi-religious cult that has grown around the UFOs. This cult, he thinks, “may be a spontaneous reaction of the subconscious to fear of the apparently insoluble political situation in the world that may lead at any moment to catastrophe. At such times eyes turn heavenwards in search of help, and miraculous forebodings of a threatening or consoling nature appear from on high.”
In ancient times, Dr. Jung explains, the UFOs might be classed as “gods,” but the unconscious of modern man has a different content. Ancient men saw dragons, other monsters and divine beings in the heavens; modern man sees mechanical portents—mysterious spaceships manned by unearthly superhuman creatures."
'...because some of these eye-witnesses were people known to Jung,[98] people he had analyzed and trained in analytical psychology, people who were at that time practicing analysts, he could not dismiss them as psychologically naïve or unaware of when they might be projecting. So what were these trusted friends and colleagues actually seeing?
This question Jung could not answer and he was left to conclude that “I am utterly unable to explain the Ufos’ physical nature. I am not even sure that it is a matter of machines; they could be anything, even animals, but I would not dare to contradict statements as to their physical reality.” '
Yeah that's a possibility. Still there are reports that we have bodies.
About FTL travel. The way they move around, like the tic tac, there is clearly some physics that we are missing. So I'm OK with something going places faster than light. Maybe they are moving through some other dimension that we don't sense so well.
I was a military aviator for over a decade, and I can promise you that the Tic Tac and similar targeting pod footage isn't remotely credible as evidence of aliens. These kinds of optical artifacts are pretty easy to create by messing with gain, contrast, and focus (especially with older-generation sensors--every single one of these videos I've seen is from the least capable targeting pod in the US inventory), and it's easy for people who don't have a deep understanding of these systems to believe nonsensical output (similar to how a video of a laser beam sweeping across the surface of the moon could be mistaken for faster-than-light motion).
Yes I find Robin Hanson's successor to his post-Fermi-paradox Great-filter framework, which he calls "Grabby Aliens" (eminently youtubeable if you have the hours to blow) intriguing. I like that it doesn't invoke faster-than-light movement nor even von Neumann probes. I happen to disagree that we would ought to be able to observe stellar-scale (or galactic-scale) engineering. Because the framework assumes that such engineering would be taking place, and that if it was that we would know what to recognize. Cosmetic point, I prefer the term NHI to Aliens, and I feel the term 'grabby' has the effect of framing perception a priori.
But I'm in great appreciation of him for opening my eyes to the prior of red dwarfs outnumbering yellow-dwarf (pardon "G-type main sequence) stars like ours by 3 or 4:1 in our galaxy, and perhaps more importantly that they are billions of years older than ours, which obvious implications for the horse race of spontaneous evolution occurring around any given star system in the galaxy.
This is an interesting update - I've personally thought there's enough evidence that there must be /something/ going on since the San Diego carrier group ("tictac") sighting was released, so it's interesting to see other heterodox thinkers coming around to this.
My favorite part of that whole incident – people often trumpet how "there couldn't be such a conspiracy, there's too many potential leakers, we haven't seen any, etc etc" – but immediately after the San Diego incident took place, before there was any public knowledge of it, some able seaman posted on Reddit about it. Leaks do happen, but skepticism is so ingrained for some at this point that they're immediately discounted as noncredible. Stuck prior type stuff.
Interesting had never heard that about the early Reddit post, would be great if there's an archive link or screenshot somewhere; But one thing I don't understand, tic-tac incident 2004, Reddit founded 2005 (??)
My basic position is that there are a lot of ways that weird clouds, balloons, aircraft, birds, unusual weather phenomena, cockpit reflections, sensor malfunctions, visual hallucinations and assorted stuff can look odd or unexplained.
It's a lot easier to spot a small blob in the sky than it is to identify exactly what it is.
There are a lot of people and sensors pointed at the sky.
Has the CIA and NSA and whatever got up to various dubious stuff. Sure.
Just because a UFO nut has access to official government stationary, that doesn't automatically make their arguments any better than the rest of the UFO nuts.
Also, lets be clear about exactly what we are claiming. Are you claiming that UFO's are aliens visiting earth? Are you claiming that UFO's have some explanation more exotic than weird clouds, but you don't know what.
And do you have direct evidence for this claim, or just evidence that someone working for the NSA believes this claim?
>Also, lets be clear about exactly what we are claiming. Are you claiming that UFO's are aliens visiting earth? Are you claiming that UFO's have some explanation more exotic than weird clouds, but you don't know what.
>And do you have direct evidence for this claim, or just evidence that someone working for the NSA believes this claim?
I don't recall making a 'claim' in the above thread? Do you feel it was incumbent on me to make a claim of something? I did make one statement (through the medium of snarkiness), that some topics in history have been stigmatized only to turn out to bear merit in studying. If that's a claim, here is my support for it: Two examples - 1. Semmelweis' antiseptic use in hospitals, and 2. covid origins (irrespective of whatever the etiology turns out to actually be)
I prepared an extended precis on my posture in a general sense w.r.t. claims but it didn't feel like an appropriate forum to provide them in earlier. With the different tone now, I can provide the following:
Speaking for only my own part and not for others who study the topic, my metagame is to restrict focus to _what some branch(es) of the government believe(s)_ rather than make any statement about whatever the phenomenon, if it exists, may or may not be. And those branches may not have disclosed their own existence over the decades since modern era of UFO observations first began.
My reason for this restriction is very mundane. When a UFO shows up and does whatever it does around people, farms, military/nuclear installations, whatever - it doesn't typewrite up a record of the encounter and drop it off into a filing cabinet for posterity. My articles need to cite things. Those things need to be documents (typically of the typewritten variety given the early postwar period I study). The only entities who typed documents in that era were humans. Readers generally don't trust witness accounts. If there were any substance to whistleblower claims of the last several years, originators of referenceable documents would necessarily be in government. So to use (@Donald's) NSA example, take its 2nd director, Lt. Col. John Samford. Maybe he says he believes, maybe he says he knows from firsthand experience, maybe he says he knows because his downstream reporting channels informed him so, maybe he simply admits in writing an intention to publicly treating the topic in a reductive threat vs. non-threat framework on the heels of the 1952 mass DC flyover wave https://www.youtube.com/watch?v=4-MbGYAv7Cg&t=41s rather than speaking to the Air Force's perception of the reality of UFOs ( as revealed by its security posture toward sighting them https://www.cia.gov/readingroom/docs/CIA-RDP90-00552R000303250002-8.pdf ). In any case, even if Samford turned out to be 'a nut', his written statements - whether he intended them to be public or not - are still citeable. I can claim nothing beyond what the documentary record itself claims. I can also speculate scenarios and frame circumstantial scenarios that are supported, but not proven by, the record. Any such statements that he makes, committed to paper with a known, acceptable chain-of-custody, would be referenceable in a research article.
That said, I don't limit myself to conclusions that can be drawn with 100% certainty from the documentary record. If the object is a government program that wants to stay hidden, then sometimes I have to make do with less evidence than academic historians (and their employers) would demand before publication. Rather, I can make the extent of uncertainty known to the reader.
(an evidence base which only grew owing to the efforts of another recently: https://majestictruth.substack.com/p/act-iii-92447-birth-of-the-machine , unfortunately the goodies being after the sub. Summary: evidence presented that DCI Roscoe Hillenkoetter (CIA's first) was present at a notorious Sep 24th 1947 meeting between Truman, Vannevar Bush, & SecDef James Forrestal. May not sound like a lot but actually kind of a big deal when all the priors are accounted for.)
I can also venture into still more speculative territory, setting up circumstantial frameworks to be poked at. A good example of that is the piece I promo'd in these threads a couple weeks ago: https://blockedepistemology.substack.com/p/there-where-uap-researchers-fear It paints a pretty clear picture of the personages & tone of the period in question. And what I think remains a pretty compelling piece of fresh evidence (note: not a 'claim' in my usage of the term, but perhaps a claim in the sense of one that can be probabilistically evaluated).
Speaking of which, that just reminded me I have a whole post illustrating my take on claim probabilities too: https://blockedepistemology.substack.com/p/belief-comes-in-two-kinds . It wasn't written for a Rationalist audience, so the style may seem coarser than folks here are used to consuming, but I think it carried the concept across and outlined where my Bayesian head was at at the time. Hope folks enjoy picking that one apart.
> I don't recall making a 'claim' in the above thread? Do you feel it was incumbent on me to make a claim of something?
I think that you have some belief, and that belief involves UFO's in some way, but you aren't being clear on what you believe about UFO's specifically.
> (p) - my assessed probability of the evidence base,
In order for a probability to make sense, you need some statement about the world attached to that probability.
When you made that expected value calculation, what events were you calculating the probability of?
>When you made that expected value calculation, what events were you calculating the probability of?
Ah, I appreciate that clarification.
below all set against the prior of the notional Drake equation background when I look at the Hubble's and then the Webb's Deep Field Image (as notionally tiled across. the celestial sphere), due consideration to young age of our star vs much older & much more numerous stars (red dwarves) in our galaxy. Parallel priors on possibility of Non-human intelligence occupying space colocated with us that we can't perceive ('interdimensional hyp'). extratempestrials...
P("fundamental UAP reality"): "g̶r̶a̶n̶d̶m̶a̶'s̶ traditional ufology" that some residuum of UAP sightings have been non-prosaic in etiology)
P("crash-retrieval") aka "c/r": that some tiny fraction of UAP engagements resulted in a crash)
set against the prior of all the witness statements that Stanton Friedman & co managed to collect, (like Marcel's & about 40 more) out of the Roswell incident, published in Crash at Corona among others.
P("reverse engineering") aka "r/e": that an unacknowledged program has existed to reverse engineer aspects of UAP flight capabilities, whether or not crashed material exists )
set against prior that there is repeated precedent for government programs to exist in secret availing of enormous resources or remaining clandestine for decades before the public would become lucky enough to know about them. (Manhattan Project, domestic spying, MKUltra)
There is a version of P("r/e") where the [attempted] reverse engineering program exists without there actually being an underlying UAP reality. Drawing a parallel to projects like Project Stargate where funds are expended on a program that relies on a part-of-government belief in a base reality that many would disagree as being operative.
The article is not one of my composition, so I don't know who at hand here you're addressing. This response may provide further context to what you ask for:
Shouldn't it at least be 'It's some *more* UFO nonsense'? or how about 'ugh more ufo nonsense from that same account that's been littering these threads with its detritus lately' ?
I'm intrigued by the timelessness & indefinite nature of the category label you selected to lump these into.
Have you considered that they might be part of the CIA? Seems awfully suspicious that they keep posting the same messages, trying to keep people from seeing the truth for themselves...
Appreciating that, and in fairness I might be having excessively lofty expectations that the audience here might resonate with the topic. I can only leave it to the author to weigh in, but I can share that generally the authors who write in this space rely on the readers' coming in with a fair amount of tacit knowledge on the history of the topic (typically the Majestic documents in particular). Otherwise the articles would be untenably long and repetitive of content available elsewhere.
I notice you still didn't mention what the whole Majestic thing is about, just "the people fomenting conspiracy theories rely on their readers being equally obsessed nutjobs".
The Fortean Times, bless their cotton socks, used to cover this and related topics heavily in the 90s/00s. For the AI précis, which is as good as anything:
"The Allegation: MJ-12 is alleged to be a top-secret committee of scientists, military leaders, and government officials formed by President Truman in 1947 to manage the recovery and investigation of alien spacecraft (specifically the Roswell incident).
Coverage: Fortean Times has explored the controversy surrounding the authenticity of the MJ-12 documents, which many researchers consider to be forged, while others (like the late Stanton Friedman) argued for their validity.
Context: The topic often appears alongside discussions of UFO crash retrievals, government secrecy, and the "all-star" cast of members supposed to be involved, including high-ranking intelligence officials"
So it's basically "little green men are real, Area 51 and Roswell are real, the government has flying saucers and is engaged in massive coverups".
By the way, funny you mention the Forteans. It's an interesting movement, yet the first time I saw them mentioned in any official context just happened to be earlier today:
"Perhaps that'll take care of the Forteans for a while."
That's Howard P. 'Bob' Robertson of 1953 Robertson Panel fame, CalTech professor, & Alsos mission veteran. 'Chad' is H. Marshall Chadwell, head of CIA's Office Scientific Intelligence of the time, which asked for the panel to convene on the UFO topic and who responded with the typed-up report of the Robertson panel:
The main practical outcome of the panel (pdf pp.23-25) was to recommend an 'education campaign' for "debunking" (scare quotes theirs) the UFO / UAP topic. The Don Marquis among those proposed (p.25) to enable the campaign was an expert academic on the psychology of propaganda (Needell, A.A. (1993), “Truth Is Our Weapon”: Project TROY, Political Warfare, and Government-Academic Relations in the National Security State*. Diplomatic History, 17: 399-420. https://doi.org/10.1111/j.1467-7709.1993.tb00588.x )
We happen to live in the timeline stemming downstream from that panel recommendation.
In my experience, most of the readers here are strong skeptics of conspiracy theories and pretty biased toward the dominant narratives of mainstream media reporting, even when there's very compelling counter-information. Examples: COVID-19 lab leak theory, hesitancy/ questions about the COVID mRNA vaccine, Epstein in general.
I would expect them to dismiss most UFO stuff as a matter of course.
For me, you'd need good evidence to convince me. Not just a few people making claims. And it's gotten harder to convince me as the ease of faking evidence has increased...because people say all sorts of things, and whether it's fame or notoriety is in the perception of the viewer.
There is plenty of evidence out there, but you need the desire to find it and read it. "UFO's and Government" is a sober start. Also check out any Gary Nolan interview.
UFO believers still don't seem to grasp what level of evidence the skeptics are asking for. Not another trove of documents, no more sworn affidavits, no more appeals to credentials. These are entirely meaningless before the magnitude of the claims.
If the claim is that there are bodies, produce the bodies and let the world's scientists examine them. If there are crashed ships, produce superior materials unknown to the scientific literature, or demonstrate their reverse-engineered death rays. If there are reptiloids governing us, then, i don't know, shoot one in the leg and let their green blood flow on national tv. That is the level of evidence that would interest me.
I'm convincible. But I'm convincible on the object level and with some effort on the part of the person who's trying to convince me, not via tactics like this:
1. Link long post that assumes lots of prior knowledge and doesn't make it clear to an outsider what the important claims are
2. Instead of accompanying the link with useful background, just hint that it is Important Forbidden Knowledge:
> Beware, some ACX readers believe that the UFO topic is "stupid". And since you know that no topic has ever been stigmatized that turned out to bear merit, then there is no need for you to click-through. (You certainly won't benefit from doing so).
3. When met with incomprehension, refuse to elaborate.
1. People on here link to technical posts all the time that necessarily assume some prior knowledge. Anything about MCP for example. (What is huggingface? What is an LLM? What is a transformer? Why is it called a transformer?...and so forth).
3. I refused to elaborate? Do you mean to Labelle's comment?
--
@Christina the StoryGirl
>weren't ideal
Perhaps not. But I also don't know how many people go and start to inform themselves to a basic extent on the topic after my original comment. Maybe as a result when I comment in the future there will be different receptivity as a result.
>to contradict
Indeed I stand contradicted.
>rarely seeing the topic come up
Yes. Wish I had a media comms specialist who could explain to me what factors induce some audiences to engage with certain topics but not others. Then I could pick my forums with better efficiency, or at least with reduced push-back.
"coming in with a fair amount of tacit knowledge on the history of the topic"
I think you mean semantic or explicit knowledge? Tacit knowledge on the history of a topic would be quite a thing to have, presumably only obtainable from having been personally involved in a large portion that history! (Sorry, I don't mean to critique your grammar, just clarifying who you think the audience for this article is.)
Can anyone explain to me why anyone would *want* to use an LLM-based AI for mass surveillance? The use cases I have seen discussed starts with the government collecting (by licit or illegal means) many terabytes of raw data, for example cell phone locations. Then it is imagined that the AI does some magic and that a dossier gets created showing that Gavin Newsome and George Soros met in Fresno at 8pm on March 3 at 3pm. But if you want to extract the Gavin Newsome info from those terabytes you use a good old fashioned algorithm to do so. Using an AI would be too slow and expensive. And if, for some obscure reason, you wanted to train an AI on all those terabytes of data, you would start with a tabula rasa and spend several months and billions of dollars. You wouldn't start with Claude. Claude was already trained on its own terabytes of data for months and $billions. It doesn't have room to learn terabytes more.
Surveillance” is the collection or acquisition of private information, but that doesn’t include commercially available information. So if you buy something, if you buy a data set of some kind and then you analyze it, that’s not necessarily surveillance under the law.
....
The reason that no one really analyzes it in the government is not so much that they can’t acquire it and do so. It’s because they don’t have the personnel. They don’t have millions and millions of people to figure out what the average person is up to.
The problem with A.I. is that A.I. gives them that infinitely scalable work force. Thus, every law can be enforced to the letter with perfect surveillance over everything. And that’s a scary future.
Even with current-level AI, it's quite feasible to just scrape all social media then say "hey Claude, analyze all data posted on reddit by acab_cop_hater, cross-reference with employment and social media data and identify their real identity, then draft a letter to their boss highlighting their problematic social media comments and initiate a mortgage fraud charge via the DoJ API."
Or, "given this high-resolution photo of this anti-ICE protest, use the ClearView API to identify all faces, determine the identity of all protesters, and set up an alert system that will notify us if they ever apply for a local, state, or federal job, or are ever subject to background search for private sector employment"
ICE is going to honkers's houses, getting out of the ICE vehicle, and waving at the honkers. Someone is clearly doing something (running plates, maybe?) Hilarious when the honkers call 911 -- for someone getting out and waving to you in your driveway??!?
It probably has nothing to do with how you or I use it, LLMs are very good at specific search functionality which military applications might need for whatever they cannot cover with basic SIGINT
More effeciency around what they already do. We already have tremendous capacity to do analysis. OpenAI and Anthropic are lowering the cost per query so to speak. So do more with fewer gpus.
The use-case that seems intuitive to me is using it as a very sophisticated filtering mechanism. When you have a ridiculous amount of data, you can't reliably find the relevant stuff with blunt tools like keyword search, but it's impossible to get human eyes on all of it (let alone the eyes of a single person who can integrate information scattered haphazardly across the dataset). With an LLM, you're (potentially) getting something approaching human-level ability to notice what is relevant, connect it to the other pieces of the puzzle, and present the higher-ups with a summary and a pointer to the relevant pieces of raw data, but with vastly superhuman speed.
I think maybe the argument is that you can use AI to piece together a narrative from circumstantial evidence. Like if you fed a bunch of info as to what a jihadist was doing for a week, the LLM can be like "yes, this behavior is consistent with someone trying to make sarin gas."
I commented that, presumably the data would be part of the training regimen, and therefore the context window limitation doesn't matter. But when you put it like that, I realize my comment was dumb and I retract my statement. The data, will indeed, probably be fed into the AI during inference.
A more practical use, I suspect, would be to have the LLM do some bulk intermediate steps so that the human analysts can spend their attention on the bits that are harder or impossible to get an LLM to do well.
For example, you might have a million intercepted phone conversation recordings. An LLM could conceivably be tasked to transcribe the recording, summarize the transcript, and tag it with some subject metadata about the contents of the conversation. I expect very little training would be needed for this, since the first two parts especially are very similar to existing LLM-based features in video conferencing and medical transcription software. The last may take some specialized training but seems like it would be in an LLM's wheelhouse to do so in a way that is directionally correct enough to be useful.
The human analysts would then run database queries to find patterns in the tags, identify conversations they want to analyze, and then read the summaries or the full transcripts depending on how much scrutiny they want to apply.
I expect they're already doing something similar to foreign surveillance intercepts and the like, but with pre-LLM speech-tp-text tools and keyword searches instead of summaries and subject tags.
Since this is pretty much precisely how AI is currently being used in mainstream social research (think oceans of text-based raw qualitative data) even as we speak, I think this is a pretty reasonable assessment.
> The human analysts would then run database queries to find patterns in the tags, identify conversations they want to analyze, and then read the summaries or the full transcripts depending on how much scrutiny they want to apply.
I would probably automate this part, too. Humans need time to think and write notes and remember what else they wanted to check... the AI just keeps going.
Yes. That makes sense. Transcription and classification of intercepted audio. A huge number of things to be analyzed, none of them requiring a large context window. Then follow up with more traditional software for matching and finding needles in haystacks.
I think we've seen enough in Iran that I can make a prediction for how it ends:
1) gradual air war drawdown within the next few weeks
2) weakened regime (probably but not certainly continuous with the current one on paper but with weakened ability to repress the people and far less weapons to threaten the west with)
3) a net overall improvement for both America and the people of Iran
4) enough problems that people who started out opposed will have something to point to and not feel forced to change their minds
I don't buy "net improvement for the people of Iran". The regime's ability to deliver *centralized* repression will be decreased, but repression is experienced locally and the local machinery of oppression will remain. Under essentially the same managers.
Except now a lot of them will be really pissed off, looking for payback against the people who just attacked their country but not being able to reach those people. And they won't be concerned by public image abroad, like the centralized regime sometimes was. And with nobody manning the "off" switch.
Military strikes on repressive military groups can be targetted. (Not discussing repressive clerics or culture, both of which can't be easily targetted). Given this, American strikes could conceivably enhance the "off" switch of "military involvement siding with the non-repressive civilians"
I'm concerned about a Libya-style fragile-state outcome. Obama called Libya the biggest mistake of his presidency, because the coalition knocked out the "bad guys" without any plan for what came next.
Obama's not being truthful about what the reasons for Libya were, if he's still talking "bad guys." I therefore cast significant doubt on his regrets (which, yes, are separate from lying his ass off about the intervention's reasons, but still! Liar should not be trusted!)
People will believe anything as soon as you offer the thinnest appeal to self-interest, even if it makes little sense on its own terms. Why would the US do airstrikes to defend the use of France's currency? Or else, why would France request airstrikes for the sake of the American petrodollar? Why did the airstrikes coincidentally happen when Libya was in the middle of a civil war? If the US cares so much about its reserve currency status, why does it wield sanctions so often in a way that creates incentives to adopt alternative currencies? If the US cares so much about its reserve currency, why is Trump trying to reduce the trade deficit that's an ~inevitable byproduct of reserve status?
My agenda? I ain't got no agenda, sure nuff. I do keep my ears open, and the folks I'm listening to have more access than "africa check" does.
The United States wants its vassal state, France, to retain control over "the francophone african states", rather than ceding them to Russia or China (I think those are the only big non-West players fishing in Africa). This is basic geopolitics.
France benefits from the petrodollar and Pax Americana (specifically, in keeping shipping open and removing pirates from the open seas). One can argue how much France actually understands this, however.
Airstrikes during a civil war? That seems like an optimal time to do them, particularly if the civil war was... shall we say, western inspired (I have no knowledge here, simply laying out a suspicion).
The United States very badly miscalculated in its sanctions on Russia. Other than that, there is no "so often" -- we've left SWIFT and the other banking institutions as a gun on the mantle, ready to be pulled when needed.
Nuland and the neocons broke Pax Americana, so I wouldn't be so certain that the petrodollar remains the reserve currency in the world. Trump's certainly not banking on America being the unipolar power that it's been -- he's putting down chips on "we need to do this to ensure there IS an America in a decade." Hence Canada, Greenland, Venezuela (draw a circle, and then write Golden Dome).
The US is already winding down the (surprisingly small) amount of support it gives Israel (at Israel's request, since they don't want the dependence either). Which was always more of a subsidy to keep defense contractors in business (useful since you don't want to have to suddenly spin them up from zero if you suddenly need them) than anything else.
US has pre-existing commitments to fork money (materiel) over to Israel in the event of an attack. Gets written into every budget, but is zero without an October Attack.
US foreign aid (not including military) is about 50-70 billion/year, (foreign military aid varies wildly but was about 10-20 billion/year total before the Ukraine spike, not counting things like nato-suppirting bases abroad or deployments in Afghanistan). Military aid to Israel is about 3.1 billion a year (and as mentioned is being wound down to zero over the next few years- there was a brief post-2023 war spike but it never approached e.g. Ukraine levels). It's just not much of an item. People love talking about it because of, uh, (((reasons))), but it was always a small piece of the foreign aid budget which is itself a rounding error in the federal budget.
I think the complaint is more that US middle east policy is dictated by Israel rather than us sending Israel a bunch of aid.
I don't understand why all those Nobel Prize winners had the brilliant idea to build their state in the world's most quarrelsome region, right next to the people they have all their historical grievances with. Even I, as a feeble non-Nobel intellect, can see that Moon Zionism is clearly superior.
It wasn't obvious in the 19th century that Arabs would be the world's most quarrelsome people in the 21st century (also, could've been much worse - imagine if they'd somehow gotten the land for it somewhere in Europe, the Germans would've just killed them all. And if they had gone with the Uganda plan... It's not like Africa is known for being peaceful. And FDR vetoed the Alaska plan. Basically no good choices)
Since their founding, we've given Israel > $300B, ~$250B of which was military assistance. So I guess if you thought we'd given them trillions, it'd be a surprise?
3) for both America and the people of Iran, things ultimately get worse on net overall
4) enough localised things being better that people who started out in favour will have something to point to and not feel forced to change their minds
My general disappointment with the state of public discourse in the post-truth era leads me to suggest that in reality the camps from your (4) and my (4) will exist simultaneously, and it will take significant work to decide which reality we are actually in for those who still care.
4) is an interesting alternative, but I think it is far less likely than you think. Iran has a co-ed dance scene, it's just underground, and for the middleclass urbanites. I'm pretty sure a lot of people who want to say "Iran is crazy" don't realize they've actually legalized prostitution.
It's good the regime is being dismantled in real time. But whether or not what follows is worse or better, no one can say. Meanwhile there is a huge human cost to bombing a country that will be glossed over by the hawks. Net good, but: be cognizant of the lives being snuffed out.
It is striking the degree to which many online American discussions of this and other wars seem completely oblivious about the fact that human beings (other than Americans) die in it. Not explicitly dismissing the value of Foreign lives, not accepting it as a fair price to pay for a greater good, just... not even recognizing it as a thing that is happening.
The War Crimes were occurring before the bombing started. Water War, led to protests. So, the "huge Human Cost" is perhaps LESS than leaving Iran alone (and letting the Israelis play water games in Afghanistan).
How do you figure? What do you think is so bad about Iran, the country that thought about getting nuclear weapons, and then decided "nope!" (around the time of the Axis of Evil.) Bibi's been saying Iran's very close for 30 years, for god's sake.
This kind of thing often ends up with a worse government replacing a bad government. And anyone anti-US will have a lot of popular support. (There are lots of ways for a government to be bad.)
It kind of makes sense. "Look what they did to us. I'll stop it from happening again, and MAKE THEM PAY!" That seems like a good sales pitch for a new regime.
Yeah, I agree with this. My expectations are something like 20% end up net worse, 30% roughly enough improvement to balance out the damage the direct strikes cause (this is lower than it would otherwise be because of how consistently the regime has supported civil wars all over the world, especially in Syria and Yemen), and 50% net improvement that's worth the cost. I don't want to gloss over the human cost here.
Unlike Libya or Afghanistan, Iran isn't a pre-state tribal society. It is in fact one of the oldest states with continuous existence, over 2500 years. They won't collapse into a squabbling bag of warlord-operated territories.
What I am afraid of is authoritarian state on steroids. The worst element of the IRGC surviving and unleashing endless hell on the population. They may have run out of ballistic missiles (or almost), but they certainly have a lot of ordinary bullets.
> They won't collapse into a squabbling bag of warlord-operated territories.
Even if Iran is reduced to rubble? I'd expect even European countries to turn to savagery if their entire government and half of their country was nuked. People act differently under desperate circumstances.
Much of Europe was reduced to rubble in 1945, just look up photos from Poland or Germany. Old nations tend to re-create new government structures quite fast, because the know-how is there.
That said, Iran won't get "reduced to rubble". It is mostly the regime infrastructure that is targeted. The US and Israel don't even have enough conventional munition to level megacities like Tehran, they would literally have to nuke them. Dumb ammo is mostly gone from Western arsenals and smart ammo is too expensive per unit to be intentionally wasted on civilian buildings with no military value.
(That does not exclude mistakes, which always happen.)
I sometimes wonder if using nukes in a situation like this would be that much worse for optics compared to leveling a place using conventional munitions, like Israel did with Gaza. Yes, a nuke is a nuke, but... I was under the impression that the newer versions were more "clean", relatively speaking. Nobody is going to start a nuclear war with the US over this.
That would still be an improvement over the IRGC! They killed more people in a single weekend than died in Libya in the last 16 years combined (and that's just domestically, not even looking at things like the IRGC-backed Yemeni or Syrian civil wars).
I think Trump fumbled when he led with his mouth instead of with strikes. Giving the Mullahs time to massacre protestors was a bad move. It looks especially bad now that we're actively engaged in the conflict. What percentage of people who were interested in regime change were killed? How many remain, and how many are not cowed into silence by the killings?
Like almost everything this administration does, this was not handled with prudence or care. A maximal show of force should not distract us from that.
The massacre of the protestors allowed identification of sympathetic military battalions. Do you want to murder all the "good troops", and leave just barehanded civilians?
Yeah. I think realistically the protests took everyone by surprise and it took time to assemble forces, but a mix of talking slower and possible early partial strikes could have made this a lot better :/
Not everyone. the people causing the war crimes were not surprised. Alternatively, the protests were a CIA thing* that just caught fire bigger than planned...
*it's possible to have locals who want to protest, as well.
1) A lot of innocent people will be slaughtered, most of them by the US, far more than the Iranian government was killing.
2) The Islamic Republic will survive. I'm very confident on this part, because history has repeatedly shown that you can't install a new government from the air and there's no public will in the US to commit anywhere near the number of ground troops they'd require. (and even if they did, it'd end up looking like Afghanistan on steroids.)
3) Oil will get significantly more expensive, with knock-on effects on economies on the world, in aggressor and neutral countries alike.
4) The war continues for much longer than Trump or the US desire, because Iran won't settle for anything less than the gulf emirates evicting US bases *and* quite understandably distrusts a country which withdrew from the prior peace agreement, assassinated their head of state, and used peace negotiations as cover for a surprise attack.
1) This is awfully vague. How many is "a lot"? A Google search indicates Iran executed 1000 to 1500+ people in 2025, which is a "significant increase in the use of the death penalty". Is your prediction that the US will kill more than 1000 civilians? It seems likely, given war, even with precision strikes.
3) Iran produces oil, but is #8 in the world (https://en.wikipedia.org/wiki/List_of_countries_by_oil_extraction) before any damage to oil production. Again, it ought to have some effect on oil prices, but "significantly" is too subjective to be proven wrong. Plus, we have no baseline for what oil prices would have been otherwise, and many things can influence oil prices.
It may be the #8 producer, but it can close the Straight of Hormuz, and significantly affect the Bab al Mandab via the Houthis. ~35% of global seaborn oil flows through those two chokepoints.
During peacetime, yes, they can certainly close it. But during wartime, I do not doubt that the US Navy can keep it open. I have already heard some things about ensuring or insuring passage, and Iran's navy was destroyed Saturday, as I understand it.
Even if the US Navy can ensure safe passage for 95% of ships transiting the straits, even a 5% risk of constructive total loss can completely upend the insurance situation. A Navy-less Iran can still credibly threaten shipping directly off its coast with cheap Shaheds that are very difficult to suppress. If insurance declares the route too risky to cover, then it's effectively closed, even if there's a good chance any given ship would make it through unscathed.
Could Iran attack a carrier in the straits, with a full normal escort complement? If not, could not a tanker, or more than one, take the role of carrier, and thus be fully protected? If so...why are these carrier groups so expensive?
I'm no expert, nor even adjacent, on military matters, but it seems sound to me.
The US has talked about providing some sort of insurance backstop.
On the other hand I would want to think about the poor crews of these tankers. A 95% chance of survival with a US-backed insurance policy might be acceptable to the oil company actuary but it's not going to be acceptable to the poor sod who has to be on board. If you tell me to sail half a million tons of crude oil through Hormuz right now then whoops I guess my oil tanker suddenly has mechanical problems which will require lengthy repairs.
Shaheds are not that difficult to suppress, if you ask the Ukrainians for expert knowledge and have as rich pockets as Arabs and Americans have.
Remember that original Shaheds are already obsolete in the Russo-Ukrainian war, precisely because they were relatively easy to suppress. The Russians are now using newer, more advanced models (Geran-2) that they produce themselves. Iran does not produce this new generation.
1. Even more people will die from the breakdown of basic services and order than from direct Israel/US bombings.
2. Islamism is still widely popular. Maybe the next leader will not be an ayatollah but he will still be a hard line islamist. Or maybe Iran will break down in a sort of feudal order with local military leaders empowered by the destruction of the national leadership.
3. Oil has already become more expensive. Plus, the worst for iranians has already happened. It is very likely that the iranians will try to block the Strait of Hormuz as there are no downsides to escalating now.
4. The aggressors keep killing the iranian leadership so there is no option for peace talks at this moment as there is no one who can lead the peace talks.
For 1), which number of people killed by the Iranian government is your baseline? The death tolls for January range from 3,000 to 36,000 according to wikipedia.
Reports on casualties in the current war are at around 1,000 at the moment, including both military and civil victims, so it's relevant which number you compare to.
However, a country having a war aim it can't achieve militarily and absolutely no faith in the negotiators on the opposite side is a recipe for a prolonged conflict.
Yeah, the thing counting against that is Trump's tendency to lose interest in things (other than tariffs) and unwillingness to stick to anything difficult.
MbS, Netanyahu, the Thanis etc. have some interest in seeing a resolution. Iran didn't help themselves by lashing out blindly on their neighbours from Azerbaijan down to Oman.
I think it will (probably) be harder for the regime to shoot thirty thousand people in a weekend again. Possibly in a patchwork way (e.g. kurds or people in some provinces will be armed, and in other places the basij will be destroyed or unable to coordinate). And I think they'll be a bit more reluctant to do massacres before compromise in the future.
For the nuclear program I think it's harder to tell - probably some setbacks but way too early to tell how much. The ballistic missile program seems severely degraded, and afaict Iran was planning to rely on it to protect them while they raced to build nukes.
(Overall I see like, 5-6 different ways in which things might improve. When I say I think this will probably be a net improvement, I think the modal scenario is something like 3-4 of those pan out, while 1-2 new bad things that I didn't list also start happening. This is pretty unsatisfactory as a prediction that can actually be graded in hindsight, but it's a broad picture of my expectations).
What are the objective criteria for “improvement”? Iran did not threaten the US in any way before this war. And it won’t threaten after. Talking about real threat not talking heads on TV.
Trying to murder Trump doesn't count as a threat? Seriously? He's not the only guy they've got on the "to be murdered" list, either (which is a way of saying "destroying military commanders is destroying operational capabilities")
It depends what you count as a threat - they've certainly been a geopolitical adversary including training and arming terrorist groups around the world and shutting down the red sea for trade last year. If you're a more hard-line isolationist that's an argument you can make (not getting into the more general argument about isolationism now), but it's at least no worse an investment than e.g. supporting Ukraine against Russia (which is far less threatening to America than Iran was).
Hezbollah does plenty of anti-american terrorism including working with cartels in Latin America. And Iran has done plenty of directly anti-american stuff (like arming anti-american terrorism in places like Iraq or Yemen, or assassinating American scientists and trying to assassinate Trump). And an Iran with a huge (potentially nuclear armed? ICBM stockpile wouldn't stop at nuking Israel.
Look, if you're an isolationist in general that's fine like I said. But Iran is at least as much of an American adversary as Russia or isis (or China for that matter), which the US invested significantly more resources into countering.
Should be easy to quantify then. Hezbollah had killed X American soldiers in the past 10 years. In the next 10 years it will be X/5. Or something like that.
You are just hand waving. Without giving any objective criteria.
Note that if you only count successful killings, and ignore the failed ones, you underestimate the costs, because preventing the unsuccessful attempts also costed you something.
Even worse, it would lead to logic like: because we were careful about them, they failed to kill anyone, therefore they are harmless, therefore we no longer need to be careful.
(This is a general note, not specifically on this topic. But I noticed how you deflected "assassinating American scientists and trying to assassinate Trump" by asking how many soldiers were killed.)
I argued with like 50 people on the Anthropic thread. One of the arguments people were making was "we're only talking about DoD/DoW use of AI for mass surveillance, no one is talking about police doing their job"
Again, the implicit argument people are making is that places like NYC have *just the right amount of enforcement* (I don't see many calling for ripping down the NYPDs camera system, which allows it to have one of the highest clearance rates in the country - and allows NYC to have some of the lowest crime rates in the country, especially given its demographic makeup), but making it any more efficient with AI would be immoral and unethical. It's a confused and nonsensical position.
The countries that use mass surveillance use a system that's nothing like Claude looking at a bunch of feeds and then making a bunch of accusations. You have cameras and ML, but it's not an LLM and it's not making any decisions.
You explicitly want this system to improve speed and massively increase the number of reports, this will inevitably lead to whatever human you put in the loop being a reverse centaur, just signing off on all the LLM's decisions, and only being there so you have someone to blame once the LLM fucks up.
Add to that that the courts simply can't handle a massive increase in cases. People would just dispute all the tickets/charges for littering or whatever en masse because they know the courts don't have the capacity to prosecute them all.
Improving speed behind a bottleneck doesn't increase the flow rate through the bottleneck and in many cases actually reduces it.
You have to explicitly make the entire system much more illiberal if you want to achieve the same speed and efficiency in prosecuting petty crime as those already illiberal places you're no doubt using as a benchmark(like China or Singapore). Just adding cameras and machine learning is not enough.
When it comes to courts and charges and clogging the system...
Just today I heard a (Czech) podcast with a realtor who sold apartments in Dubai. He commented on the local surveillance culture: "If you do as little as showing someone your middle finger in public, you will receive a fine within 15 minutes on your phone."
The punishment mechanism can be automated as well, though it is possible that the very distinct US judicial tradition would not allow for that.
>The punishment mechanism can be automated as well, though it is possible that the very distinct US judicial tradition would not allow for that.
It absolutely can. If you don't care about small things like right to appeal.
There's some argument to be made that the societal gains from increased public order would trump the losses from injust punishments, especially if we're only talking mild fines, but the US system is absolutely not set up to do this.
It would take a massive societal and judicial reform to get it there, my point was that just letting an LLM sift through camera feeds would be of little benefit and might just make things worse.
I am not endorsing that way of punishing people, I am just pointing out that this is possible.
The US criminal justice system is already quite heavy tilted toward plea bargaining, to the degree that the original jury system only applies in 1-2 per cent of all criminal cases (IIRC). This looks like an "optimization" of a similar sort, and this happened gradually, without a Day D which would mark a major reform.
I think my concern is specifically that the people in power will abuse this new system to attack their political enemies. With an AI to scale it, they could (for example) make a rule that anyone who probably voted against them in the previous election gets extra-strict enforcement for littering and jaywalking. This might not be difficult to notice but it would be very difficult to prove; the government would say "look, they totally did it, we have all this evidence!"
Trump is already doing some stuff like this on the state level; I searched for "Trump cuts aid to blue states" and found a bunch of articles. If we gave him the ability to attack blue voters individually, I think he would use it. And, to be clear, I wouldn't trust the Democrats with power like this either.
Downthread, you've argued: "It would be very good for the country to move toward fewer rules that are more consistently enforced." I agree! But I'm worried that this AI stuff will take us in the opposite direction from that.
Adding a second line of thought in a second comment for clarity:
There's a difference between 'Surveillance cameras in public spaces have been good for stopping crime, therefore we should add more cameras' and 'Surveillance cameras in public spaces have been good for stopping crime, therefore we should have an AI processing all the camera feeds and acting as an oracle for police.'
AFAIK, police don't randomly review security cameras all day, randomly searching for anything that might be a crime. Instead, when a crime gets reported, they look at the recordings from the area to see what happened, or look at current feeds once they have a suspect they're looking for.
More cameras would give them more coverage to do those activities more effectively.
But adding an agentic AI that can process all the feeds at once and put together queries, databases, etc., is a qualitatively different thing.
For example: an AI monitoring every feed can tag every litterer in the city, but the police don't have the manpower to prosecute every litterer in the city; nor every jaywalker, every speeder, every person who parks an inch too far from the curve, etc. Instead, they end up with a database of minor offenses for every citizen in the city, and can decide at their own discretion who to prosecute or not, can have a list of offenses they can hold over anyone they want to go after or who challenges the m or displeases them, etc.
Mayube yu think that's good too, maybe you think we'd reform things such that this capability was used responsibly and led to good outcomes.
But my point is that this is a qualitatively different set of capabilities than what modern security cameras grant the police. Therefore, 'cameras have been good, therefore AI will be good' is a non sequitur.
> For example: an AI monitoring every feed can tag every litterer in the city, but the police don't have the manpower to prosecute every litterer in the city; nor every jaywalker, every speeder, every person who parks an inch too far from the curve, etc. Instead, they end up with a database of minor offenses for every citizen in the city, and can decide at their own discretion who to prosecute or not, can have a list of offenses they can hold over anyone they want to go after or who challenges the m or displeases them, etc.
How hard can it be to go after every speeder/illegal parker? Lots of countries do this: There's a tired civilian cop that gets a new photograph every minute on his or her screen, validates that it looks what the description says it looks like, clicks the big green button, the owner of the car is sent a payment link to pay the fine. Sixty fines an hour, 450 fines every working day. If the car owner wants to dispute the fine, they are free to do so.
It's harder to do this with litterers/jaywalkers, because facial/gait recognition is much more subjective. Modern neural networks can be surprisingly good, but the human in the loop won't be able to confirm their guess, not unless the AI does something like: "Here's Ms. Doe leaving her house in a green jacket and sunglasses, photo courtesy of her Ring. Here's a receipt for an ice cream, paid be her card at 12:50. Here's a woman in a green jacket and sunglasses, same height, same hairstyle as Ms. Doe, eating an ice cream, throwing an ice cream wrapper into a bin and missing at 12:52. Here's her path through the city, courtesy of her phone, that places the phone in the vicinity of the bin at 12:52. Here's the list of all other people seen in the vicinity of the bin at 12:50-12:55, all of them have a negative match with the litterer."
The first problem with this approach is that now a cop needs at least five minutes per fine to issue it. 90 fines a day is still nothing to scoff at. The second one is that now that standard of evidence is lower. "Black BMW, license plate 2FAST4U" is a fact that is hard to challenge when captured by a speed camera. With Ms. Doe we have to learn to evaluate and accept this kind of circumstantial evidence that *excludes* other perpetrators, Holmes-style, instead of fingering her directly. And be we I mean the judge and the jury, because this *will* be challenged. If not by Ms. Doe, then by Ms. Chang or Ms. Washington.
I don't agree with you. The answer isn't to give up on crime enforcement, it is to actually fully enforce every law on the book.
"For example: an AI monitoring every feed can tag every litterer in the city, but the police don't have the manpower to prosecute every litterer in the city; nor every jaywalker, every speeder, every person who parks an inch too far from the curve, etc."
If there is a law on the books that you don't agree with then take it off the books. I agree that discretion is bad. Your response is to say "then we just shouldn't monitor crime". Mine is: no, we should definitely monitor crime, but use technology to enforce everything that is on the books. If you don't want something to be a crime, then take it off the books. In a just society, someone who litters, tags graffiti, etc gets a nice juicy fine in the mail the next day. And if you think that isn't fair then make graffiti or littering legal.
Your preferred system is that we should just have it be luck based who gets punished (a cop happens to be nearby)? That's crazy!
I don't think it's physically possible to enforce every crime, including minor crimes such as littering or jaywalking, without automating the enforcement along with detection; and detecting crimes while lacking the ability to enforce them is a waste of time and money.
Currently, the criminalization of e.g. jaywalking acts as a deterrent. On the one hand, when you jaywalk there's a small chance you'll be stopped by a cop; but on the other hand, jaywalking can serve as a factor in the prosecution and sentencing of other crimes. For example, if you get run over by a car, the driver would normally be found liable -- but if the traffic cam video shows you jaywalking, then his liability would be reduced, potentially down to zero. Due to these factors, some people who would normally choose to jaywalk abstain from doing so.
Merely automating enforcement won't get you very far, unless you fully commit. For example, you could stick a camera on every corner and automatically mail out tickets to jaywalkers. Some will choose to pay the tickets, but most will probably ignore them (this is what often happens with speeding cameras in the real world). The next step is to summon the alleged criminals to court, but in order for that to happen, your courts need to have the capacity to try all those cases, which they currently do not (they barely have enough capacity for speeding tickets). As the result, most jaywalking cases will remain unpunished.
The next step is to automate the courts as well, or bypass them altogether: if the jaywalking AI deems you guilty, then you're guilty, and refusal to pay the fee makes you increasingly more guilty until the police are dispatched to your house to put you in jail. This could work as an effective deterrent, assuming you put in enough effort in the initial phase of your crime-fighting campaign to stage some very public arrests. That said, the cops don't have enough manpower for a sustained effort like this ether, but perhaps you could optimize them by putting the AI in charge of dispatching cops as well.
And now, you live in a paradise where almost no one ever jaywalks; in fact, people are rarely leave their homes at all, for fear of being randomly thrown in jail by an unaccountable LLM with no due process and no hope of exoneration. Like I said... paradise !
You have a distinctive profile image, so I recognize you from some of your other comments.
Frankly, I think that most of your comments have been snarky, glib, and dismissive. Most of the people who you've responded to seem to share my opinion.
In contrast, I think that this most recent comment of yours was articulate, fully explained, and polite.
In case you found yourself putting substantially more time or effort into this comment than your other ones, I want to let you know that the difference is both noticeable and appreciated.
To be honest, when I post my comments I don't set myself a goal of being snarky vs. articulate; rather, I always strive to convey my honest opinion on topics that I find interesting. Sometimes I express agreement, sometimes I express disagreement, and sometimes I express strong disagreement indeed -- it all depends on the topic. That said, I'm only human, and as such I've got to admit that I am pleased by and thankful for your attention !
That's because Singapore is a small country that actually does have the resources to enforce all of its laws (with no AI automation); and Shanghai is a city in a country that is fully onboard with implementing the police state and all the joys that come with it. I don't know what the situation is like in Singapore, but in Shanghai, the average citizen is in constant danger from his own government -- much more so than the average American is from jaywalkers.
This is the same thing as the SAVE act: I'll support requiring ID to vote *after* we have a competent bureaucracy in place with a 10 year track records of making sure every citizen has valid ID or can get it in a few days with under an hour of effort. But what they're offering is to require an ID to vote *while* the bureaucracy is a complete mess.
Sure, I am absolutely in favor of reducing the number of laws and regulations, and the schedule of punishments, down to a level where I would be happy with 100% perfect enforcement. And after the legal system looked like that, I'd be happy to give the police the tools they need to reach 100% perfect enforcement.
But you're advocating doing the second step first. And I'm saying that will lead to horror and dysfunction.
I too wish we were already in utopia and could act accordingly. But I won't apologize for recognizing the world we're *actually* in, and acting accordingly for that world.
When you succeed in reforming the entire legal system into a perfect ideal that can support 100% enforcement, get back to me and I'll join your side.
Assume that 50% of Americans no longer trust when a Democrat gets elected. Do your priors change now? The ID - requirement can be as performative as the TSA (literally never passing inspection), and still be good for America.
No, you don't reward liars by giving them the thing they're lying to get. Especially not when it not only incentivizes, but also directly empowers them to keep lying more in the future.
As they say, the decision theory on this is straightforward.
So, it's your assertion that you actually know about the ballots, and have crystal clear evidence, contra the boxes carried out of Fulton County by the FBI? That seems rather speculative on your part. I wouldn't sit around and say "I know that election was fair" -- not when we have more than one party* that would rather bollox the election, on general principle, and more particularly because of one Donald Trump, existential threat to the Middle Kingdom.
One of the things AI is changing is that a whole bunch of processes were naturally bottle-necked by human attention, and they were calibrated to have that limiting factor, and now that factor is gone.
"Well, duh, fix those" okay but we have lots of things to fix. Breaking it to find out what we need to fix is going to cause lots of problems and we already have a big list of broken things already.
I'll say the same thing I said in that thread - surveilling public spaces with cameras, and surveilling private correspondences and actions through data mining, are not the same thing.
I don't actually know how much public surveillance NYC does and how much it actually helps with crime, I definitely don't naively trust anything the NYPD claims, if I did a deep dive I might want more or less than the current level, I don't have a strong prediction.
But that's really not what we're talking about here. I think it's bad for people and bad for the country if every action you ever take on or near a digital device in the privacy of your own home or place of business is being watched by the government. Adding an agentic intelligence that can actually process all of that information at once and act as an oracle for government agents makes it much worse.
Public surveillance sucks, but you were already being surveilled in public by the other people in public, you already had to steel yourself and put on a public face and be on your best behavior until you made it from one private space to the next.
Being under surveillance *always*, having *no such hting* as a private space *ever*, is bad.
Bad for the soul, bad for society and culture, bad for our relationship to the government.
The point of the Panopticon as a thought experiment is that the Panopticon is bad. When you are forced to build a policeman inside your own head and perform the actions of innocence and compliance for it at every moment of the day, you eliminate the space needed for your own individuality and humanity to flourish.
Trust me, you do not actually want to live in the Panopticon. It is not fun and not worth it.
You are exactly who I was referring to! There is no evidence whatsoever that Anthropic is exclusively saying "don't use the tech to data mine everyone's digital devices" - the other article lays out fairly conclusively that they just don't want their tech to be used to uphold the law.
This is a classic position I see from a lot of Western leftists, when you don't agree with a law (or are uncomfortable with it for whatever reason), instead of changing it, just throw as many barriers in the way of being able to faithfully uphold it.
Anthropic's position is implicitly that police and related law enforcement are the only careers in the country which aren't allowed to have productivity increases through technology.
>There is no evidence whatsoever that Anthropic is exclusively saying "don't use the tech to data mine everyone's digital devices"
Who cares about 'exclusively'?
If one of the 5 things they are doing is horrible, I don't want them to do the horrible thing!
> instead of changing it, just throw as many barriers in the way of being able to faithfully uphold it.
First of all, I'd love to change the laws I disagree with! But I don't have unilateral power to change the law whenever I want!
There are plenty of activist groups trying to reform the actual law, and I support them. But given that they're not close to success on a wide variety of topics, of course I also support other efforts to mitigate the harm caused by bad laws. Why would I limit my efforts only to things that won't work?
Second of all: the legal system in the US has never been designed with the intent of enforcing all laws at all times. That's an insane thing to think.
Between jaywalking, littering, speeding, intellectual property, a 2600 page federal tax code, different buildings codes by states and county, byzantine contract laws for every different industry, and everything else, there's probably not a single adult citizen of the US that has never broken a single law or regulation.
Perfect enforcement would put everyone in jail, including the police.
Selective enforcement has always been the law of the land, and this has always meant that the police have discretionary power to wage cultural and political wars against groups they see as targets and enemies.
Giving them more power to notice small crimes and generate evidence doesn't lead to better enforcement, because enforcement was already limited by how much the police feel like enforcing and how much manpower and time they have to do it (plus court availability etc). All it does is give them more discretion about who to prosecute, and more leverage over anyone who protests.
It would be very good for the country to move toward fewer rules that are more consistently enforced. In fact that is what separates a lot of developed countries from developing countries.
While I see your point in general, I don't see why the better solution isn't just "get another contractor to do that". There's lots of other AI companies that are basically almost as good (and will be better in six months than anthropic is now).
Yup. If you just compare homicide rate for each ethnic group individually, NYC is near/at the lowest.
Out of 64 major counties:
Black homicide rate per 100k: Queens = 9.7 (2nd lowest in the country), NY County (Manhattan) = 13.0 (7th lowest), Kings County = 13.2 (9th lowest), Bronx = 16.3 (13th lowest)
> Again, the implicit argument people are making is that places like NYC have *just the right amount of enforcement* [...] but making it any more efficient with AI would be immoral and unethical. It's a confused and nonsensical position.
I disagree with this interpretation.
By analogy, this is like saying that if you support the US Military, you would want them to have the ability to kill anyone in the world instantly, without cost or consequence to themselves. But if they had this ability, would they kill the exact same set of people that they do now? Or would they also be tempted to do way worse things? Maybe the President would decide to kill every member of the Iranian government and the IRGC and anybody else the CIA said was bad. Maybe the threshold for targeting people would go down to a lower level of consequence. Maybe somebody would ask why we don't just kill everybody who looks at us funny.
Obviously AI doesn't let the NYPD vaporize shoplifters but it's still fair to wonder whether it might have some other impact on their behavior if ticketing people or surveiling people became much more trivial.
Also you're just completely ignoring the possibility that hallucinations at the scale of a 10M person city could add up to a lot of issues.
I think we just live in completely different universes if you think that NYC has even close to the optimal amount of punishment for crime. Anti-social behavior carries virtually no penalty. I could fare-evade, shoplift, litter, tag a building, etc. literally every day for the rest of the year and there would be 95%++ probability that I would not face a single consequence.
And that's NYC, with relatively high clearance rates (due to cameras!). The clearance rate in other U.S. cities is abysmal.
> I think we just live in completely different universes if you think that NYC has even close to the optimal amount of punishment for crime.
Which is not actually what I said. My point is that, one, giving somebody more capabilities doesn't just result in them doing exactly the same thing except more or better, and two, I have concerns about using unreliable systems in a criminal-justice context.
Law enforcement isn't just, like, one dial for "amount of punishment" that has a perfect setting that people are too weak to choose. There's a bunch of tradeoffs. Given that, as you yourself keep noting, NYC has a high clearance rate and low crime, but also people sometimes object to specific things the NYPD has done, I don't think it's surprising that people feel like maybe we shouldn't give them a tool that will probably cause SOMETHING to go up when we can't totally predict what that thing will be.
High clearance rate *relative* to other U.S. cities.
My test is very simple: can a resident of a city bike to their local corner store / bodega and go inside while leaving their bike unlocked. If the answer is "no" that means we are below the optimal level of punishment.
I dunno. The reason your bike won't be stolen in Japan isn't because the police are very good at catching bike criminals. It's because the people have lower propensity to steal bikes.
You could argue that crime clearance rate can decrease the propensity for crime but up to a certain point.
While you have a certain admirable consistency about you here, I feel like you're not really engaging with me on the question of whether there are any tradeoffs, or whether reliability issues are a concern in how we deploy these systems, or even more generally the question over whether this is why people oppose it rather than merely a belief that crime is punished too much or too often.
And I've stated before but I think that the reason most people opposite it because being against surveillance is a secular religion.
If it was so terrible then westerners who moved to Singapore would be doing everything in their power to try to dismantle the system. Instead, they forget about it and enjoy the fact that there is zero crime (without ever thinking how).
Vitalik is a great example, he obtained permanent residency in Singapore and lived there for years. But now he is against the U.S. having the same type of safety and order that he thought was valuable enough to get residency for.
It's actually very simple for me: I dislike disorder and crime so much that I am willing to give up a lot to see it go down. Imagine if you were in charge of Venezuela's economy and someone said "there might be tradeoffs" to liberalizing it and undergoing market reforms. The current status quo is so horrifically bad that you would take that risk.
A number of my highly intelligent clients can be off-putting. These tactics have been helpful:
Consciously decide when to lift the gas pedal. We all like to be our best, but sometimes, that imposes too big a price. So stay aware of when the risk/reward might be better if you held back, just as a pilot pulls back on the throttle when it’s worth sacrificing some speed for safety, as in a cloud bank.
Yes, it’s cliche, but do listen carefully to your conversation partner(s.) Assess how complex your utterances or writing should be. I often err on the side of simplicity. Even highly intelligent people like to understand without strain. If you’re unsure how to respond, whether in content or level, you might ask a question. Not only does that clarify that issue, it makes the other person feel validated, as long as the question is not asked condescendingly. So, it’s usually fine to ask, “I’m not sure I understand. Could you paraphrase that?” but not, “I don’t get it. Try again.”
My clients and I benefit from, as appropriate, imagining explaining a concept to a bright 6th grader. Often, that’s complex enough. You can always add complexity, but it’s often better to at least start with that as baseline. I have tried to do that here.
Catch them doing something good. Even if they never say something you deem brilliant, if it’s something you think worthy, at least nod and smile if not say something like, “Good point” or “Good job. Thank you.”
Soloing. Of course, when you’re soloing, you needn’t worry about dialing back your brain. So, even if you’re social, spend enough time on solo activities, for example, those that build on your strengths. Possible examples, writing, performing, even gardening.
The latter offers perhaps surprising opportunities for intelligent optimization. For example, I hybridize roses. First, I thought through the best niche — Generalist breeders fail unless they have a large-scale operation. So I put myself in the target customers’ shoes: older people who like the romance of roses but require a small plant because they don’t want or can’t dig deep holes or live in an apartment with only a window box to plant a rose, and they won’t spray with chemicals. Then, I carefully selected varieties to cross. I’ve never taken even a course in plant genetics, but a Google search and $20 membership in the Rose Hybridizers Association, taught me how to choose parents, even which to choose. Then, I created a highly efficient system — including a way to minimize the time-consuming but critical recordkeeping system. I have two roses on the market and nine under test with top U.S. rose distributors.
Kindreds. You’ll be more effective and likely happier when you’re with other high-cognition people. Today, that’s a more acceptable term than “intelligent” or “high-IQ.” Because we spend so much time at work, it helps if your coworkers are cognitive peers. Say, you want to work on Google moonshots. Might you know someone who could introduce you to someone there? Like most highly selective organizations, the applicants that get interviews have usually been referred by trusted colleagues. If not, trying sending a brief, compelling email to decisionmakers, even if a job isn’t posted. The subject line should make clear you’re more likely to be signal than noise. For example, if I were looking for the aforementioned job, my subject line or first line would be “Berkeley Ph.D. in evaluation of innovation with testable ideas for increasing LLMs’ user agency while better optimizing safety vs utility.” Most unsolicited emails get screened based on the subject line.
Outside of work, consider certain MeetUps and online groups, for example, in STEM, debate, or chess. A word about Mensa, where I’ve been a member. I have been disappointed on two counts: It’s more a social club than a think tank for addressing societal problems. And Mensa seems overrepresented by underachievers. I am pleased to see, however, Mensa making special efforts to recruit promising young people — a shout-out to Lana Lee, who is doing great work on this.
Curate your helping. In or outside of work, your help is often most effective when the cognitive gap isn’t large. Counterexample: The large gap between professors’ and students’ cognitive ability inhibits student learning.
Go wide and deep. Most people face a lose-lose choice. If they dabble, they risk being a jack of all trades, master of none. That can be fun but the lack of depth can impede career. On the other hand, in going deep, they too often reach a point where greater depth is needed for career success, but that’s too challenging. That’s a fatal flaw of science museums and of introductory college courses. Those offer enjoyable basics, which hook people, but when they get to, for example, the higher math required in most STEM careers, they quit, having incurred significant opportunity cost. A benefit of being quite intelligent is that you may derive the benefits of going wide and deep.
Forgive yourself. Even high-powered minds screw up and need breaks. I know someone who, to decompress, sews boxer shorts for her boyfriend while watching cartoons. Another pulls weeds. Another reads books for young adults, like The Giver.
Is there one, maybe two things, on this list you want to try, do more of or less of?
The macro
At the societal level, I’d like to see:
Improved restoration of K-12 programs for gifted kids and honors programs at universities. They’ve largely been abandoned in favor of programs for special-needs students. The “best and brightest” are the most likely to cure cancer, better address energy issues, and yes, reduce wars. Yet today, too much of their potential is going unrealized. Programs for the gifted should not be limited to advanced cognitive work but include human issues, like those addressed here.
Media pivot. In earlier generations, excellence was more often venerated by the media and higher education. Today, the plight of low achievers is emphasized and, in some circles, cognitive excellence is viewed as elitist, even racist. Restoring balance is key to long-term improvement in gross world flourishing.
I read this aloud on YouTube.
Marty Nemko is a career coach specializing in people with high cognitive ability. He is the author of 36 books including Work Stories. He holds a Ph.D. from Berkeley with dual specializations: educational psychology and the evaluation of innovation. You can reach him at mnemko@gmail.com
This is missing an introduction that establishes context. It can't capture my attention because for the first several paragraphs I'm just thinking, what is this even about?
This is all great advice, and thanks for posting it.
I'd add one thing you didn't address at all:
There are major returns to doing something physical and pushing at least 20% on the "physical optimization" side of things vs being all-in on the mental and intellectual side. Get into a sport, or running, or swimming, or lifting, or some sort of regular physical activity.
Your physical health affects everything, including the quality of your thought, how long you can sustain rigorous thinking, your overall vigor and energy, and of course your morbidity and all cause mortality.
Being sedentary vs exercising regularly is a worse all cause mortality effect than smoking, and being obese has additional ACM increments on top of that.
If you think of the "elements of a life well lived" I think they'd be something like this:
* Good health
* A sharp mind
* An interesting life, full of curiosity and complexity
* Time spent working on problems that fully engage your strengths and are towards some meaningful goal
* Physical vigor and robustness
* Fine meals, expertly prepared, with the freshest and highest quality ingredients
* Robust social lives characterized by truly reciprocal love and devotion and mutual desires for well being and success
* Thriving kids and grandkids
Having the physical dialed in helps most of those! Being physically engaged and adept improves nearly all of those things.
I think there are many issues on both sides of the political isle. However, to be honest, it seems like one side (conservative) is either kept from all the facts (echo chamber) or is openly ok with bad behaviors, policy, constitutional concerns etc in a “ends justify the means” sort of way. It also feels that the “policy” of the right is hard to keep track of. I heard so much from that side about not starting other wars and and “American first” doctrine, yet now that it seems we are doing the opposite, such voices seem to be silenced? I’m trying to approach this with an open mind as I know I have my bias. Background, I was staunchly conservative until Trump and MAGA policies began to shift GOP conservative. Now non registered but leans left and often votes left (but I take it case by case, person by person when I vote). What am I missing. I’d love to hear from conservative voices here in a spirit of understanding.
An awful lot of Americans really, really hate the Iranian regime. Forty-plus years of "Death to America!" rallies have left an impression. And if there are some leftists in the space between Canada and Mexico who aren't terribly fond of "America", those are the ones for whom Iran beating women to death for showing their face in public will be the dealbreaker.
And I'm pretty sure the vast majority of Americans understand that Donald Trump is not going to be impeached for starting this war any time soon.
So, as long as it can be imagined that this will be a Short Victorious War(tm), the pushback will be muted. If you ask people about it, many of them will (and do) note that it was probably stupid and illegal for Trump to start this war without Congressional approval, but they'll also say that they hate the Iranian regime and want to see it fall. And they won't go out of their way to raise a fuss, because what good would that do?
They all get to watch the hated Iranian regime take a hammering, and they can believe that they'll get to see the regime fall, and the ones on the left may imagine they'll also see Trump impeached later on or the GOP take a drubbing in the midterms or whatever. None of that requires anything of them now, so the mostly aren't doing anything now.
When it becomes clear that this isn't going to be a Short Victorious War, that either the Iranian regime will remain in power or that we're being sucked into a Vietnam/Iraq-style quagmire, that may change. But for now, Americans are going to be pretty chill.
You dont think hes impeached by the house if dems take it?
Im not sure how this plays out. I think its objectively bad that he has now had two long planned attacks evidenced by months of military build up, with out even bothering to try to get an authorization for use of force. GOP controls congress so he would have gotten it, and it would have if anything strengthened his negotiation hand in each case.
Its true that the average american would basically be fine with it if both cases turn out fine. But if for other reason his approval is below 30 in 2027, its clearly a good candidate for the legal simplicity of the case.
As for his own reasons, I wonder. It could be cmon no one bothers with these, Obama didnt. Im just going to do what im going to do. It could be the more cunning let them get used to me just blatantly violating the constitution so i can do so again in ways that arent just about process formalities.
He may well be impeached by the House next year. That does not require action or even comment by any protester or dissident *now*, so they're not going to say or do anything now that might harsh the vibe of all the people around them who are just enjoying the show and dreaming of the fall of the Iranian regime.
There is no single right wing block. The older republicans that have supported Bush's Iraq war also support this one, while many on the new right hate this war but don't want to turn on Trump.
The strikes against Iran are not what people in America generally use "war" to mean, and I think they're consistent with how the "no new wars" campaign slogan was understood by his audience, whose central examples are WWII, Korea, Vietnam, Iraq, and Afghanistan. No _invasions_ is perhaps a better translation. Drone and missile strikes are fine, as long as you can credibly claim not to be motivated by "democracy," "human rights," "stopping genocide" or whatever, which are the kinds of things liable to get you to put troops on the ground and start an occupation to ensure those goals are achieved.
I think this is what Hegseth tried to convey with "This is not a so-called regime change war, but the regime sure did change, and the world is better off for it."
How is this different from the NATO strikes on Libya, which Trump called a "total catastrophe?" Iran has 12.5 times the population of Libya and is closer to Europe. If it turns into civil war there's a real risk of another refugee crisis.
There's an ocean between us. Turning away boats or sinking them is much easier than protecting a land border. Boats are much more visible than people, especially ones that contain a lot of people.
Iran is not closer to Europe than Libya. Libya is in fact quite a short sail from Italy, which is one of the reasons why it became such a people smuggling hub.
Well, Iran has a land border with Armenia, which is usually considered a part of Europe, despite being located to the east of the Asian part of Turkey, so that would make the distance from Iran to Europe 0 km, while the distance from Libya to Greece and to Italy appears to be somewhere between 200 km and 300 km.
But I doubt that worry about Iranian refugees pouring into Europe is focused on Armenia.
And that's exactly what happened. The entire NATO Libya operation was seven months and no boots were put on the ground. The US involvement was just 12 days of bombing.
Except the regime hasn't changed, just the guy at the top. Now his son is in charge of the exact same regime.
The Iranians seem savvy enough to read between the lines. The Trump admin's policy is, "Short bombing campaigns project strength and play well to the base, let's do lots of those. Extended campaigns lose support over time and we never win them, let's avoid those."
Last June, the Iranians agreed to the short strike option as a way to get the US off their back for a while. That didn't work, so they changed tactics. Now they have lots of options to escalate this war regionally, and they seem unafraid to do so, because they know that plays against Trump's stated strategy. Meanwhile, the Israelis also want to draw the US deeper into the conflict, as they always have. They're not about to allow the US to bow out after a few weeks of strikes, especially with how long they've been trying to get the US into this thing in the first place.
Sure, quick strikes seem like a viable political strategy, until the enemy hits back and digs in for a prolonged fight. Time will tell, but it looks like Trump has made a profound miscalculation here, thinking he could pull what he did in Venezuela, or do another Midnight Hammer. At the very least, Americans are dying. That's not supposed to happen in the "commit acts of war, but no new wars" strategy. His best bet is to find a quick offramp here, but he no longer controls whether the other two major players in this conflict will let him keep his policy.
That sounds...incredible. Do you have any evidence for this interpretation? I'm looking at current polls, and no matter what you may call it, Americans don't seem to find these strikes "fine".
I'm basing my assessment on previous strikes, such as on Soleimani last term or the Houthis last year, which people took in stride. You're right that the polls don't look good, but I contend that they're using the question as a proxy for "do you think this will escalate to an invasion – an 'actual' war - and would you support that war?" They get yippy anytime Trump does anything, but I expect that's a transient thing which will pass in a few weeks when/if it becomes clear it's not going to become "World War III" or whatever the current hysteria is.
(Of course, there's also a more secular decline in his approval from his "Après moi, le déluge" economic policy, which also colors some of this polling, and is bound to give the Democrats a trifecta unless he drastically changes course. His foreign policy, whatever it is, isn't going to meaningfully affect that either way.)
>I contend that they're using the question as a proxy for "do you think this will escalate to an invasion – an 'actual' war - and would you support that war?"
You are correct, strikes can lead to escalation, invasion. Therefore, if Republicans voted for Trump because they want "no more invasions", shouldn't they poll with much _lower_ approval for strikes than Democrats?
I think this has a lot to do with the state of online discourse where we are bombarded by ever more sophisticated bots and then have genuine people who are plugged into think tank talking points.
I almost voted for Trump, and part of the reason would have been the anti war rhetoric. Particularly from Vance and Gabbard. And here we are, we let Israel drag us into this war. Which does nothing for us. WTF. And the weird thing to me is all the MSM seem to be in support of the war. Is there anyone in congress against it? Rand Paul? Why this almost universal support in our congress for Israel?
Congressional Democrats are like "while we fully agree Iran is really really bad and (list of horrible things khomenei has done) Trump is not explaining adequately why he's doing this, what the plan is, and also isn't Congress supposed to be the branch of government that declares war???" Source- an email I got from my congressional representative to the constituent email list. So no media filter there. Sample size of one, but I think mine is considered a fairly standard center of party Democrat.
I don't think Israel dragged America in. Israel could have and would have done this on their own if they decided it was necessary and saw an opportunity, if America hadn't been ready they wouldn't have waited for them. (Proof: last June.). My pet theory is that this is somehow connected to Trump's personal financial interests in a way that will probably be exposed at some point in the future.
OK the Dems can and will say all these bad things about Trump, and he has no plan and didn't ask congress*. And still support the war. IDK.
My take on Israel comes from the breaking points podcast. Where I say a clip of Marco Rubio saying (roughly), "Yeah we knew Israel was going to strike Iran and we knew Iran would respond against the USA, so we went in there preemptively and took out there missiles and such. So they couldn't hurt Americans." That counts as 'dragging us in' in my book. I don't think it's some Trump evil plan, but that's just me.
*As I continue to say the biggest prob I see with the way our government works is that congress has ceded too much of it's power to the Prez. If they wanted to take back that power they could. And control the war and the budget and tariffs and... They are wimps and we need better congress people.
It's difficult to understand these justifications. If Iran was planning to attack the US first (even if Israel attacked them first, we could at least claim we weren't fighting Iran at the time) that would have given the US a casus belli for war. It's an absurd response to the question, "What's your justification for this war?"
"They were going to give us one, but we beat them to it and gave them an excuse to retaliate whole depriving ourselves of a reason to fight." This is a far cry from Lincoln, setting up the South to attack Sumpter - which they obliged him - and give him a reason to mobilize the public against the aggressors.
The NYT headlines on the war have been mostly negative, focusing on the spreading side effects of the war and the fact that Trump doesn't seem to have a clear justification or goal for it.
They were never going to be for it. This is the NYT.
Trump could physically charge into a burning building, walk out with an armful of kittens and babies he saved from most certain death, and the NYT would spin it as a negative.
This is about my feeling as well (bias warning: started as a cross pressure moderate, drifted left over ~10yrs to be a classical left-liberal).
In general both parties seem at their worst when bowing to their ideologues (Trump for the right, progressive tastemakers for the left). But the centralization of power on the right seems to have legitimately excluded the conservatives that I could imagine making common ground with, while I think much of the mainstream Democratic Party still has a lot to offer people of all political persuasion.
I will agree with Shaked K. in this thread that Blue State (city?) governance produces a real image issue for the left at the national scale.*
*A little out of scope for this comment but I think that the realities of national politics tends to curb the worse impulses of the left in a way that’s not symmetric to the right.
About your asterisk, I think the bigger curb isn't just the realities of national politics, it's that it's easier to get competent high level administrators for big programs run by Washington than for state or local issues.
(That said, mostly I just don't see what the Democratic party has to offer me anymore. I used to like industrial policy or infrastructure projects or Medicare, but the major programs for that all seem so consumed by waste and fraud I've changed my mind and would prefer tax cuts now).
I've heard from people on twitter (el gato malo) that there's open war in the GOP netizens, in terms of isolationists versus interventionists versus Trumpian-uber-alles supporters.
This...does not reflect my understanding of Republican politics.
To start, there are pretty deep fractures in the Republican base over this. You can see Tucker Carlson condemning the Iran war here (1) and you can see Ben Shapiro taking some pretty pointed shots at Carlson here (2). I think a lot of political actors are beginning to prepare for the post-Trump era. He's old, obviously slowing down, and the post-Trump presidential campaigns should begin after the midterms. It's very unclear where the party will go though; there's no unanimity among the pundits, politicians, or candidates.
If the Republican party seems contradictory on this, it's because internal factions within the party are fighting over narrative control.
I'm in the exact opposite boat from you (lifelong Democrat until 2025) and I have the opposite impression (for example, most people I've talked to on the left seem sure that Milei destroyed the Argentinian economy, and every supposed moderate democrat immediately jumps to the left five minutes after being elected).
It's easy to point at bad things Trump does (I probably agree with you on most of them). But I still lean right overall, because I think the left would be significantly worse in power. We don't see it partly because they're not currently in power (and extrapolating from Obama or even Biden to current Democrats would be like extrapolating from Romney to Republicans), and partly because Democrats' mistakes are more invisible - going too far in letting process or bloat accumulate is generally a bigger issue than acting too severely, but it doesn't cause sudden problems so it mostly gets overlooked.
We also just have an empirical comparison between red and blue states - I think it's pretty hard to argue that California or New York are well governed compared to Texas and Florida (you get similar levels of service at best with significantly higher taxes).
None of this is to say there aren't real problems with the current version of the Republican party. But I think it's more salvageable than Democrats, so I prefer trying that instead of going over.
The current administration attempted to enact the largest tax hike of my lifetime, and the only complaint coming from the party about it is that the Supreme Court stopped him from unilaterally creating a massive tax hike by Presidential diktat. Come on man.
I agree that the tarriffs are very bad in multiple ways, and are a prime example of the ways in which Trump is often pointlessly destructive. That said, Republicans are still broadly speaking the lower-tax party (especially at the state level). More importantly, they're still broadly deregulatory (Trump has a few big pro-regulation areas about which he is very bad, but he also put the 2/10 for 1 cancelled regulations per new regulation rule, which makes up for a whole lot of bad stuff).
Going forward, I think the best political strategy is to try to empower the traditional/Rubio wing of the party over the populist/maga pro-tarriff wing. It's not that I disagree about the destructive decisions Trump makes (again, I opposed him in all three of his elections!), I just think it's unfair to compare him to a neutral placeholder instead of the likely alternative.
> That said, Republicans are still broadly speaking the lower-tax party (especially at the state level).
Again, quite literally, the largest tax increase of my lifetime was enacted by Donald Trump, and approximately 0% of the Republican Party or its base pushed back on him about it. You cannot say you are anti-tax at that point, I'm sorry, I don't believe you. You had the most obviously bad tax policy imaginable, enacted without appropriate mechanisms (e.g. Congress writing a bill), and they just bent over and took it. It's fucking tax policy! It's not some weird deregulation of pharmaceutical laws or something, "don't take my money" is the most vulgar form of conservatism there is! It's the conservative equivalent of lefties complaining about billionaires!
It's over, MAGA won, you can't cite 40 years of effective libertarian-leaning policy when the current leader of the party has a meltdown about a Canadian ad regarding Reagan opposing tariffs and everybody in the party just goes along with him!
> I just think it's unfair to compare him to a neutral placeholder instead of the likely alternative.
The "likely alternative" is Vance, not Rubio, who is also, err, not pro-deregulation.
e: Sorry, wait, if you meant likely alternative as in Biden or Harris or Newsom... bro! Not a single Dem in my entire life has raised taxes as much as this motherfucker, come on!
The current administration is an incompetent mess. Just like every other Trump venture. He’s just breaking shit. What has he done that will have a positive long term impact? I can point to him defunding and destroying immensely valuable institutions endlessly.
There's some obvious things, like the ongoing Iran campaign or various deregulation measures, but those are controversial so let's put them aside for here. I'd argue that (a) some amount of creative destruction of institutions is necessary (to quote Elon Musk, if you don't regret at least 10% of the things you removed you're not removing enough things), but mostly just (b), I think Trump is bad but less bad than a typical Democrat would be. If we could go back to the Bill Clinton Democrats - something you can draw as a normal baseline - I'd take that over Trump, who really does as you say do a lot of stupid or destructive things. But the current energy in the democratic party is on far more destructive left-wing policies than that (you can look at how the places where Democrats are in power to get a sense of the results), so I now think Trump is less bad than they would be.
(Note that this is new for me - I didn't think this way in the 2020 or even 2024 elections, and I agree with most criticisms of Trump. But seeing the Democrats' reaction over the last year has pushed me over the edge).
> some amount of creative destruction of institutions is necessary
It's only _creative_ destruction if you actually _create_ something, otherwise it's just _destruction_. Trump has, in just one year, done more damage to America's soft power and reputation as a reliable economic and military partner than any other president managed to do since the end of WWII, by a wide margin.
> to quote Elon Musk, if you don't regret at least 10% of the things you removed you're not removing enough things
Some actions can't simply be reverted in the next software release. Once you've openly and explicitly threatened to invade an allied country for non-sensical, petty reasons, there's no going back. Sending Rubio with a superficially soothing speech to Munich won't undo the damage you caused.
First off, this is wrong. Creative destruction happens through better systems displacing and replacing existing ones, not by deleting the existing systems and then figuring out how to fill the resulting vacuum.
Second, Trump isn't exactly the kind of guy who has the experience, the will, the attention span, or the consistency for the long, arduous, and unglamorous process of building new, competent institutions, nor does anyone in his administration [1]. It just doesn't make for "good television", and burning bridges or humiliating partners is much less boring.
[1] Maaaaaybe Rubio does. But as long as he has to bend to the daily whims of his boss, he won't get the chance to do so.
What about the outright grift and corruption? The tariffs which were paid for by USA consumers? You think we should have eliminated waste in the government. For the deficit? Then why did we give tax breaks to the billionaires? You are aware their effective tax rate is lower than any other citizens?
Did you catch the Democratic elected officials trying to invoke "secret hidden signals" to countermand Trump's position as Commander in Chief? One of them got his rank/pension busted down by the DoD. That's far from invisible.
(skim down to where Senator Elissa Slotkin starts talking. It's in her words.) You don't need to look at el gato malo's philosophy to look at their own words. "Don't give up the ship" is weirdly put, and strangely put. The whole thing reeks of coded messages (have you seen the video where Nancy Pelosi's brainwashing gets triggered? It's hilarious -- she literally reboots).
What? How "credulous" do I need to be to see them court-jumping from "let's prosecute this in civvie courts" to "let's use the UCMJ and the JAG lawyers" to "ya know, let's just bust him down a rank and call him slapped down"?
Many popular conservative pundits and influencers will say whatever they’re paid to say or will go along with the party no matter what. They aren’t principled and just serve a function of herding the masses. Trump has betrayed his base on virtually every issue and some supporters are slowly starting to realize that.
That said, it’s not like the left is somehow more principled. They never cared about the lawfare against Trump, the denial of due process for J6 offenders, illegal censorship during the 2020 election, and countless other issues. They are no more well informed - take the ridiculous claims made during the Kyle Rittenhouse case for instance. Or basically anything related to woke ideology.
Without relitigating the entire thing, there were all sorts of nonsense takes going around about it somehow mattering that his home town was on the other side of a state border, or that he was a murderer or a terrorist of some kind. In reality there was abundant video evidence of it being textbook self defense.
J6 folks sat in prison for so long that their sentences were commuted, in general. 6 months? free to go, you already served that before the trial.
Folks are rightly upset if they want to complain about "this took too long" -- this was "bill of rights" level egregious conduct by the Biden Administration.
That's not "denial of due process". The criminal justice system is indeed too slow, and people will take plea bargains because it's quicker, but that LONG predates J6. J6ers getting commuted by Trump means the deviation from normal process was in their favor.
(In general I think the fact that we take years to do trials instead of days or weeks is a pretty big travesty. I don't know if it's specific to j6 people though? Maybe if the government didn't let any of them plea out for political reasons)
sorry, I should have said jail rather than prison (there is a difference), my bad.
In general, it doesn't take years to do criminal trials (and when it does, it's because someone wants it to take that long, often the defense -- if Chauvin wanted a longer trial, I wouldn't blame him -- police in prison have a rough and short lifespan). These were excessive, to the point that the process was part of the punishment.
I honestly think both sides have approximately equal percentages of low informed voters and ends justify the means voters, but there is an important difference between the parties. The Republicans have many more politicians that are low informed or ends justify the means
Does this mean you should stop being a Republican? It means the opposite, at least if you’re in a closed party state. You should stay a Republican and vote in primaries for candidates that are well informed and do not believe the ends justify the means. Democrats primaries are not nearly as important as Republican primaries. You can still vote Democrat in the general if need be, but you are an important force in making the Republican Party back on track. At least that’s been my plan for the last decade.
Are you talking the economic terrorist Republicans? Because I think we pretty much got rid of them, and the new economic terrorists are on the Democratic side.
*Yep. "economic terrorist" was bait, read out by an Irish-American Representative. When folks tried to call him on it, he mentioned that he was just quoting the former Republican Secretary of the Treasury (a constituent).
Steve Yegge just announced Wasteland, which is his federated version of Gas Town, which in turn is the most aggressive multi-agent coding tool I've ever seen. It doesn't stop at writing code, it will merge PRs to main, by default.
This is exciting for me personally since the backend is now running on Dolt, so we're seeing tons of new usage for the database and DoltHub. But in terms of whether the idea of federated multi-agent AI rigs working together is insane, or useful, or doesn't go nearly far enough, I am very uncertain. I'm curious what ACX readers make of it, especially if you've tried Gas Town yourself.
I'm not nearly rich enough to try GasTown. I feel like you have to be some kind of eccentric, successfully-exited founder to be willing to flush the amount of tokens really running it requires. For my scale of projects, an interactive Claude Code session is plenty productive. Frankly I think GasTown is going to quickly become an obsolete curiousity of an intermediate age, and all functionality it provides will ultimately be subsumed by the models and model providers themselves. You can already see this coming with Claude Code swarm mode, Codex multi-agent management, etc.
edit: I see swarm mode is officially "agent teams" now that it's out of stealth
I barely even use Claude Code and I still constantly hit the usage cap ($20/mo sub). I really wonder how all the hypesters and vibecoders are doing it. Are they really all paying 200+ a month?!
I've read all the Gas Town blogs and now the Wasteland one (thanks for the link, BTW!). I have not used Gas Town and am not likely to, although I am a heavy Claude Code user.
Gas Town has some interesting concepts (its reliance on Beads for logging issues discovered during implementation for one, although Beads wasn't working well for me on Windows when I tried it a month or so back) but to me it's a very Ptolemaic system -- epicycle upon epicycle to try to keep the underlying models coherent and on task indefinitely.
Does it work? Maybe. I will be more convinced when I see significant software produced using it that is not itself or its own tools. For now, my personal sweet spot is setting up automation workflows with Claude Code that involve up front detailed specs with human review gate checks. This approach uses my human judgment and knowledge to keep the workflow moving in the desired direction rather than a crosstalking collection of agents roleplaying a corporate development structure.
I don't know if my way is more effective, but it's simpler and more scalable for what I'm using it for.
Don't forget Discord (an ugly sound, dissension and strife), Palantir (a corrupted tool of the great Enemy), and nVidia (invidious - offensive and unfair, tending to bring resentment)
I read through the Gas Town post but have not used it. I don't think I can justify spending money on it out of morbid curiosity.
When I read about these sorts of things, I often think the same thing, which is that I have a general theory of what my job is, and apparently it's very different than Yegge's theory of the same.
Most reasonably-complex software systems are much too large to fit into your brain at once, unless you etch them into your long-term memory. Most of my job is to deeply understand systems so I can know what is reasonable to do with them.
If I no longer have to read through the code a bunch to make changes, I feel like that's getting efficiency today at the cost of efficiency tomorrow (or forever) by taking away my understanding of the system.
I'm sure some people will say "You don't need to understand the system - the LLM can read the whole thing and understand it in seconds", but that doesn't really stop me from getting the LLM to make terrible mistakes because I ask it to (and cannot recognize them as mistakes during review).
Also, I write a disproportionate amount of the incident reports for my team (mostly out of, again, morbid curiosity) and a lot of things that cause incidents are quite subtle or even arguably-sort-of-not-wrong.
I still think these are useful tools. I use them regularly. But they give me wrong answers a lot, even with the latest models. And they struggle when I give them more complex tasks (my current attempt is along the lines of "find me any code that isn't technically dead but seems like it really shouldn't get hit ever", which seems to be too much because in most cases you need domain knowledge to appreciate what "shouldn't" happen).
Thank you for this. I have been reading Yegge on and off since "The Kingdom of Nouns" and have a strong bias to believe he knows what he is talking about more than I do but this is a well reasoned counterpoint that matches my own experience.
I'm still confused about what is actually being built. As far as I can figure out from the Wanted board, Wasteland is mostly just building Wasteland so that it can build Wasteland?
Yeah, the tendencies towards LARP-ing, gamified badges, and "recursive-self-improvement" have crowded out anything else this system might do.
Also: he says 1500 pull requests is "a lot more than most companies have done in the last 2 months" - but I managed 350 in February with a normal usage of Claude Code / ChatGPT Codex.
I have not tried Gas Town, excited to hear about Wasteland and always interested in what Yegge is doing.
I finally got around to using Claude Code over the last few days and...I'm pretty ok with letting this thing run loose. I think it will make mistakes, I think there are types of tasks I wouldn't turn over to it but...it's not significantly dumber than me or my coworkers. It was like...I basically ran it in another monitor while I was in meetings all day, checked each code change as it was made, didn't see anything wildly stupid, and approved it. That feels "safe" and I'll probably do it for the next 6 months but...man, I feel really unnecessary in this loop.
I can totally buy a fleet of coding agents doing crazy work.
ACX readers, what is your current AI setup, accounting for a wide range of use cases and risks? So many recommendations, so much confusion. Sample idea from my colleague:
Level 1, top privacy: locally run models for highly sensitive stuff (e.g., Llama 8B + LLM Studio)
Level 2, higher privacy: OpenRouter + pay-per-use models, email/payment method/prompts without personally identifying information. Alternatively, Perplexity subscription for convenience.
Level 3, low privacy: Claude Opus and Gemini coming with Google Workspace for public-facing work, email/payment method/prompts with identifying information.
(Plus, if you’ve overshared in the past - non-obvious data poisoning)
None whatsoever, though I'm considering an Anthropic subscription if Dario holds firm.
I can't use any of the public models for my work, because everything is CUI or proprietary. The few times I've tried to use the company's approved in-house model, the results have been worse than useless except occasionally for entertainment value.
And I haven't found any really compelling use cases in my personal life, at least from what I've seen from colleagues who have tried it. An amusing toy, but too unreliable to be trusted with anything important and I don't need it for the unimportant stuff.
Also, I have a philosophical objection to sending money to Sam Altman, and I'd have to think real hard about Elon Musk in this context.
I have a $20/month Claude subscription, a $20/month Google AI Pro subscription, and a ChatGPT Pro subscription I got for free. I tend to use them like so:
- Gemini for things that involve visual reasoning eg, validating structural design questions, complex multistep identication-and-reasoning from images, etc. Nano banana (2) is also very good at image generation, on the rare occasions I want that.
- Claude for coding related queries, sometimes in the web UI and sometimes Claude Code, both via the terminal version and the vscode extension. And the occasional fun query when I want "someone" to bounce a strange idea off of.
- ChatGPT for the occasional quick query, ie the name for a specific part, help recalling cases I can't remember the specifics of, etc, plus the occasional Deep Research report. I found the 5.1 personality so annoying it soured me on the whole brand, even after I updated the default. Apparently 5.3 is supposed to fix this, but Claude is just nicer to interact with, overall.
Almost daily use for tech work, via API pay as you go, what you'd probably call level 2. "Review this code", "how would you implement <whatever>", "explain this error message", "translate this sentence", etc. It has mostly replaced StackOverflow and online manuals for me. Just getting into more automated tools like Amp code and Claude Code. I also use the API to ask general curiosity questions about anything from Sanskrit etymology to elderly relative's medications.
On the phone, random questions go to the free tier of the chatgpt or gemini apps (level 3 in your list).
OTOH, I regularly turn off "AI help" features in apps, browser and OS's. I want AI to answer when I ask, not to do shit on its own.
At work, I use the Copilot plugin for VS Code. My employer's IT and legal departments have signed off on us using that, and I figure it's their call whether it's secure enough for the company's purposes. I make relatively light use of it, but do find it somewhat useful for intellisense-like purposes.
I was about to get a new Mac for Level 1, but my Apple Store has cleverly stopped selling Macs temporarily until the new ones come out in a week or so. The man is always trying to get me down.
Do ya'll think Iran executed all the protestors from last month they had sitting in prison? On the one hand, I haven't heard anything to indicate this has happened... on the other, it seems like a really obvious move during an attempted coup (and what they might have been planning anyways).
The legal opinions around the OpenAI contract are interesting to read, but there's one very fundamental issue that no amount of lawyering can reassure, right?
The crux of the original conflict was that the DoW insisted upon the right to cross at least one of Anthropic's two red lines.
No matter what assurances anyone from OpenAI or DoW gives, it is unquestionably the DoW's understanding that the OpenAI contract language is more permissive than their contract with anthropic would have been. Otherwise why they bother with this whole rigamarole? Am I... Somehow mistaken on this? I feel like this is the point that needs to be pressed within OpenAI.
If Sam Altman believes that the red lines as would be understood and interpreted by a normal human off the street are in place, while the DoW is nonetheless satisfied, then he is asserting that he is somehow hoodwinking the DoW in plain sight, no? Doesn't pass my smell test.
IMO Sam Altman knows that the red lines have been crossed, the DoW knows that the red lines have been crossed, OpenAI is okay with losing a few customers over this, and OpenAI is just making empty statements and doing security theater to try not to lose their employees. Dario Amodei is a biased source but I really agreed with his memo
I think Sam Altman is talking out of both sides of his mouth here, he pledged to stick to the same red lines as Anthropic AND grabbed a bunch of Pentagon business, all within 24 hours. I think he's just saying whatever sounds good, like many other CEOs.
It's most likely that the new agreement was due to some perceptual or contractual loosening, but it is possible that the DoW straight up chooses to treat Anthropic and OpenAI differently because OpenAI has more political favor with the admin and the reverse for anthropic.
I think you're probably right, although I also wouldn't put it past the current administration to be so stupid and narcissistic that they would switch just so it looked like they "won" and/or be tricked by OpenAI.
Either OpenAI's contract let's the DoW do more stuff than Anthropic's or it doesn't.
If it does, well, that explains everything.
If it doesn't, DoW might still have gone for it due to all the donations OpenAI people have given to Trump for example, and more generally the ways in which OpenAI has sucked up to them.
Both possibilities seem plausible to me.
Another important piece of context is that Sam can just say one thing to the public and a different thing to the DoW behind closed doors, e.g. "Don't worry all this talk of red lines is just for show, we won't complain if you quietly change the terms of this contract later and/or blatantly violate it"
I agree with both this comment and Yug Gnirob's comment below.
It's either that the contract is more permissive, or there's some other explanation for why this all happened (likely from the set of options presented in the replies.). In either case, public reassurances about the quality of the safety stack or legal language are noise when it comes to resolving this core concern.
I myself don't have a strong feel for which one of the offered up answers is the real one, so I'll refrain from making any guesses here. Just that the official story being presented in public does not add up no matter how you cut it.
You're assuming the DoW is following their stated logic, instead of looking for an excuse to preserve their ego. They're the kind of people who would take a worse deal so they could lie and say they made a better one.
Tariffs are a point in which "consistency" is very clear. It's just that the agenda is very different than you think. Chaos is a means to an end, a very particular end.
Deflating the stock market on Trump's terms. It's in a massive AI induced bubble, and if Trump deflates it, Trump can reflate it as he pleases (a managed recovery if you will).
It's a high stakes ploy (and there's other agendas at work, too), but Trump kinda likes those.
It depends on whether Trump puts boots on the ground. IRG says it intends to fight to the last man. I think the Pentagon wouldn't want to invade. Trump made a spontaneous Truth Social post the other denying that we're running out of missiles. Funny thing is, no one had brought this up as a possibility. So I have to wonder if we aren't running out of missiles?
I don't think there will be a ground war in Iran. If we run out of missiles, and Iran runs out of missiles, it will probably fall into a stalemate. Israel may keep it going, though.
I'm very new to this, but a few random factors that could prolong things on the Iranian side:
1. US is supporting the Kurds, who also get support from Iraqi Kurdistan - there aren't enough militants to actually threaten a major takeover, but having an active Kurdish insurgency in the Northwest could keep some kind of conflict ongoing for a long time.
2. Random other insurgencies like the People's Fighters Front/Balochistan stuff.
3. Khamenei succession might end up being chaotic - probably some messy internal politics that it's difficult even for an expert to understand, let alone a noob armchair observer. Khamenei's son seems like the likely successor, but the irony of a revolutionary republic founded on overthrowing a hereditary monarchy doing father-to-son succession will surely raise some thick Persian eyebrows!
4. The IRGC-adjacent Basij militia seem to be a factor that stops the "fairytale" idea of a liberal uprising - a big chunk of them are pro-regime thugs who are embedded in society... but I've got no idea how deep their loyalties run
5. The Artesh (conventional army) is likely to gain power relative to the IRGC, as the strikes continue to decimate IRGC infrastructure and leadership. This might mess up domestic power structures a bit, and give more power to the (less extreme) Artesh. Can't see this leading to a coup or anything, but having two non-aligned militaries could make Iran a lot less stable.
This is all true, but there's an alternative story, where Shia Islam makes a play in the various countries allied to the US, or nearby, and the conflict spreads out of control. We don't know what the next Supreme Leader of Iran will do, but with the protests around the region this week, there seems to be a lot of potential unrest that could go against the US presence. Lebanon has joined the fight. The Houthis have vowed to shut down oil exports in their region if the Saudis join the war.
The US strategy seems to be that if they create hardship, in an unpopular regime, ruled by autocratic dictators, who are a minority presiding over a silent majority that has a long-standing hatred of the government - that this combination of factors will be enough to destabilize the regime and lead to a popular uprising.
Okay, but that describes every US ally in the region, only moreso. And now none of those regimes are getting oil income.
Proxy groups like Hezbollah and the Houthis survive because Iran funds and supplies them. The Houthis are probably not building anti-shipping drones themselves. It's hard to know how effective these groups will be if Iranian support becomes unreliable.
Seems clear from what we've seen over the past year that it's a more nuanced relationship than a strict proxy fight. Iran pulling back would certainly hinder the Houthis, though I don't think it's fair to say Iran totally calls the shots of what they do or don't do.
Meanwhile, China seems interested in using this conflict to their benefit. As does Russia. I'm not keen on discovering what creative surprises they're currently working on to make life more difficult for the US in that region.
Many countries, China, India Japan South Korea have limited stock of oil and rely on continuous import from the Gulf. They are going to start yelping pretty soon.
Probably a few weeks for the main bombing campaign, with some long term low level warfare (e.g. things like arming the kurds). The navy, main missile factories, and much of the governing infrastructure has been taken out ahead of schedule Afaict, without a land invasion that's probably close to a full air war victory.
I doubt it will go on for long. When it's a liability domestically, the Trump administration will change it's tune. American casualties will accelerate the process. It's a terrible time, politically speaking, to be getting American soldiers killed in a war of choice. And this was a choice: Iran was no threat to civilians residing in North America.
So far American casualties are very low (less than Pakistan's!), because we've been avoiding "boots on the ground", and Iran's capacity to strike outside its borders has been so diminished.
The US wanting out won't be enough to get out, unless it's also willing to withdraw significantly from the region
Iran correctly sees the US bases in the region as a significant threat and is probably unwilling to settle for "okay we'll stop so we can do it again in a few years"
I can't see the US pulling out if Israel is still fighting. Bibi has wanted the US in this war for 40 years (his words). He's not about to let us walk away after a couple weeks of bombing. They dragged us into this, they're not letting us back out. There will be escalation, it's just a question of what that looks like.
Iran's government is a threat to civilians residing in Iran, where tens of thousands were recently murdered. It doesn't look good complaining over a few American deaths.
I'm highly skeptical of this claim. What verification do we have that it's legitimate? It seems to have come from unnamed Israeli sources, but I'm open to better sourcing if you have it.
Do you realize how logistically difficult it is to kill that many people in so short a time? And how difficult it would then be to verify that number that seems to have come from nowhere? This looks exactly like the kind of talking point that has been used for decades to get people enthused for every other war in the ME. Can we please not fall for it again this time?
What were know from that period is:
1. There were lots of protests (both pro/anti regime)
2. There were professional agitators among the protestors
3. Those agitators were armed with fully automatic weapons and targeted police and firefighters
4. The Israelis admitted they were involved
What this looks like is that there was an attempted color revolution that the regime violently suppressed. It looks like lots of innocent people probably got caught between these two sides fighting for power. Legitimate numbers I've seen are in the low to mid thousands. That doesn't sound like a regime being careful with the lives of its citizens, but like panicked people doing what they can to stay in power.
I'm not here claiming the Iranians have some great track record of human rights. But they didn't wake up one day and decide to start slaughtering people days after Trump publicly declared that one of his red lines for intervention was if the regime started mass killing their own people.
This was never about the humanitarian situation. You know what because the US is allied to the Saudis, among others. The US entering this war will almost certainly cause more Iranian deaths than the fake tens of thousands being used as a pretext for war. Just like in every other conflict where this tactic has been employed.
Official estimates from the Iranian government are over 3,000; various human rights activist groups claim to have verified 5,800–7,000 deaths; other estimates are higher. "Tens of thousands" is sufficiently disputed that I'll stop using that, but "ten thousand" does not seem to be an exaggeration.
10k vs 20k does not change the moral calculus.
The motivations of Trump are, ultimately, irrelevant to the moral question here. I'm pleased just to know that most of the people who made the decision to murder those people are now dead.
I'm not certain about that. Let's say 8k, which is on the high end of the credible sources I've seen. Of those, I've never seen a breakdown of how many were killed by the regime versus the terrorist groups gunning down the police. Is it 50/50, 90/10, 10/90? There's no answer to this other than vibes/bias at the moment. Partly because the regime isn't interested in letting in human rights groups or free speech. That's not cool, but then again the two countries that just openly started an aggressive war against them bragged about sending terrorists into their country to kill policemen. My country has done far more with far less justification .[EDIT: here I'm referring to the speech restrictions, not the killing of protesters; the US is far behind in that category, with a much smaller - though nonzero - body count].
Whatever the balance, I doubt this intervention will cost fewer lives on balance than the status quo. I don't expect the nation that sent terrorists to kill firefighters in Iran to be careful about civilian casualties. I've seen the videos of mass bombing over Tehran. Reportedly they're planning to send gravity bombs next. Not exactly targeted strikes.
Meanwhile, my government lies to me. All the time. Never more consequentially than when they're lying us into war. Vietnam (Gulf of Tonkin), Iraq war 1 (babies in ovens), Iraq war 2 (WMD), Syria (chemical weapons), and on and on. Later we find out that they used our compassion to get their war. Large numbers of people die so the US can ride in on a white horse chasing windmills.
I'm out. Call me an isolationist, but better that than following these warmongers who always find some new excuse, but never an actual cause for war.
Calling the Gaza campaign genocide is an ugly smear. It should stop. The entire city had to be razed because the entire city is in service to terrorists. The violence is terrible, but appeasement and doing nothing did not work. October 7th proved that. It was time for Gaza to go.
I'm fed up with this ridiculous pretense that Hamas is a threat to Israel. Hamas is a pathetic third world mob with kalashnikovs and fireworks. It should not be a threat. The rational thing to do, even just economically let alone morally, would have been to build a figurative wall around Gaza and otherwise let them be.
October 7 was carried out by a mob with small arms, which would have been stopped by something as cheap and simple as a minefield. Somehow it took Israeli helicopters half a day to take off. Israel ignored an explicit warning by Egypt on October 4 that a big attack was about to happen, as well as many other warning signs.
The lessons Israel should have learned are: Set up a minefield and other simple defenses. Don't ignore warnings. Make sure that helicopters take off immediately.
Instead, they went with מָוֶת לָעֲרָבִים
I assume that the reason Israel did not take precautions was that they did not care. They knew that if Hamas tried a large scale attack, they'd respond with genocidal destruction, which would make whatever government would carry it out popular with the Israeli electorate.
Has Hamas been destroyed yet? No, they're still there. What is stopping them from doing another October 7 tomorrow? The fact that Israel keeps an eye on and defends the border. That is the active ingredient that prevents another October 7. Not genocidal fury, but alertness. And you know it's true because nobody thinks another October 7 is possible tomorrow, or back in 2023 on October 8 or 9 or 10.
Genociding Gaza has cost Israel well over 100 billion dollars. With one tenth of that money you could have built defenses that Hamas would never be able to get past. For the security of Israel, each dollar spent setting up landmines and concertina wire throughout a Korea-style corridor around Gaza would be worth, I don't know, 10, 100, 1000 dollars spent genociding Gaza. Not that Netanyahu cares. His goal is not the security of Israel, his goal is to satisfy the bloodlust of his electorate, and to make it possible for settlers to steal land.
"Calling the Gaza campaign genocide is an ugly smear"
"It was time for Gaza to go"
yeah, it was genocide, you just support what was done to those poor people
And Israel did not try appeasement, what it did beforehand was a perpetual siege with occasional October 7th style murder raids. They never offered real independence or control of their own borders.
Anyone with any understanding of military reality knows that if Israel wanted to kill everyone in Gaza, they could get most of it done in an afternoon.
Instead they fought an urban war for 2 years and only killed 3% of the population.
The genocide accusations can only come from deep ignorance and/or passionate antisemitism.
America is not the world police. Terrible things happen all over the globe. There is nothing particular about Iran that makes it morally righteous for us to get involved. Iran is a belligerent, but they are a threat primarily to Israel and other countries in the region. They do not have the technological or military prowess to pose a threat to anyone living in the geographical United States. I am excluding military bases from that calculus, though those are treated as being US soil. Almost every single American casualty is due to us having a military presence in the region.
When you get right down to it, this is about our ally Israel. Iran is not shy about it's desire to see Israel gone. We are supporting them in ending a threat to their existence. When it comes to America, they are an old man yelling at a cloud. But their cruise missiles can physically reach Israel. They bankroll terrorists that kill Israelis. I am not opposed to the air strikes, but I would like to see it end there. This is Israel's problem, not ours.
This to me is definitely the quote of the day when it comes to the Iran war.
“is very complete, pretty much.”
God that made me laugh
My fear is that the young and impressionable will come to think of it as a norm. Oy.
Thoughts on Pelicot mass rape case: the reason I do not understand it at all, is that men are competitive. And seducing women is an achievement, a trophy, raping unconscious women is not, it does not give one more bragging rights than an inflatable sex doll - rather a definite "desperate loser" vibe.
So I don't get it. Definitely not simply patriarchy, because the competitive attitude to sex is also patriarchy. Perhaps a particularly vile, hateful subset of patriarchy.
Epstein rapists are somewhat more relatable, because getting into that elite exclusive circles, a secret society, is an achievement. But ultimately I still don't understand that, because one can just party on that island without raping teenagers and it is the same achievement. Also, getting into a secret society does not count as a big achievement by the likes of Clinton, Trump or Chopra, those people can get in everywhere.
I think you can't explain urges rationally; it's like asking why people eat sugar when they know that it makes them fat, sick, and unattractive.
Similarly, if someone feels a sexual urge and there is an opportunity... and it is a powerful person who is not used to getting "no" for an answer... and it's guaranteed by other powerful people to be unpunished...
And there is no dilemma. You can rape teenagers from poor families on Epstein's island on Monday, *and* seduce the ladies on Tuesday. (And the dark truth is that most of the ladies won't even mind, because they don't perceive the poor teenagers as their equals.)
Men are competitive, but sometimes they also share with friends. And some of them have weird fetishes. The Pelicot case seems to be about a fetish, the Epstein case is more mainstream.
Of course you can explain urges rationally. How thirsty do you need to get before you're willing to drink -- "insert unpalatable beverage here"? The Male Sex Drive, left unslaked, is a higher strength urge than thirst, which is already a 9/10 (Male sex drive being a 9.5/10).
Explaining Epstein as "urges that don't get punished" is like trying to tell me that f*cking goats is also an "urge that doesn't get punished." It's a poor model. You'd do better conflating Epstein's island with noble cannibalism.
How could you possibly understand it? You are clearly not the sort of person who is a rapist. If you did truly understand it, you would almost certainly feel ashamed of yourself. That said, it’s a pretty strange case.
Intelligent people build models of others, that enhance understanding. If you aren't doing such, your life is probably in danger, due to sheer ignorance (not, in particular, from rapists, necessarily.)
Building a model is one thing and I agree with you we need to do that. Understanding something I take as a different order of knowledge.
Understanding someone/something is being able to predict it, in novel situations. Either you understand electricity or you do not (yes, there are gradations, not understanding QED might mean you can't do certain things, but you can certainly still plug in a lightbulb).
Other people do not necessarily emulate the laws of physics and chemistry. You cannot compare understanding electricity to understanding other human being.
Emulation isn't necessary to model other humans. In fact, it may be counterproductive, since emulation means you're "thinking like them" and thus you don't need to learn the models from scratch. Learning to build models of humans is a difficult, lengthy process.
First they came for the Communists, but I did not speak out -- because I was not a Communist.
Then they came for the socialists, but I did not speak out -- because I was not a socialist.
Then they came for the great clown Pagliacci. I burst into tears. Said, "but, doctor, I AM Pagliacci".
This is brilliant! Thanks.
Fox News showed videos of Ukrainian drones fighting Shahed-type drones, and pretended that it was the latest American AI-powered technology.
https://xcancel.com/wilendhornets/status/2029543527533674600
I am not surprised by the part where Fox News lies; obviously. But I wouldn't predict that Americans will show someone else's weapons and pretend that they are theirs. That's something I would expect from someone like North Korea, who *wants* to be seen as powerful, but *isn't*.
They played the wrong "drone footage" clip – the guy who was talking, Brett Velicovic, was describing the strikes against Iran's leadership, while the videos were of Iranian drones being hit – and fixed it in a later segment, with rather effusive praise for the Ukrainian technology. Even for Fox News critics, you're scraping the bottom of the barrel here.
https://xcancel.com/zoya_shu/status/2030393006281679358
It's definitely nice of them that they fixed it after they got caught.
https://boriquagato.substack.com/p/the-narrative-bombs
More adventures in stupid media.
Murray Gell-Mann was right, don't trust the media, it never knows what it's talking about. Obvious mistake, possible misleading by government actors, actual treachery, it doesn't matter. These guys aren't correct 10% of the time, and when they are correct, you could have observed it walking down the street.
Current "hysteria in media" for example is about oil, which is like screaming at the slowly moving elephant for being big and gray, while the cheetah is actually mauling your child. Expecting the mainstream media* to actually do their job is a step too far, apparently.
*as opposed to trade publications, who publish Actual News because it's the only thing that's within their remit.
So I am posting my small problem amidst these discussions of world events and big issues: I would like to be able to "Like" posts here. GPT said to install TamperMonkey in Chrome, and then install a script that it wrote. Did that, but the script is not working. When I come here, I still don't get an option to heart a post. GPT thinks the script is OK, and probably I need info from someone here about what works with the present Substack to unhide the hearts option. Can someone point me in the right direction?
FYI, I don't like getting the rare 'like' for a comment. If you like what I say, just write that as a comment and say a little about why you like it. (Or don't like it...)
This comment simply infuriates me. ;)
You'll be happy to hear I gave him a heart for it.
Gave you one too.
The Chrome extension ACX Tweaks has an option to enable likes.
Thank you. That worked.
You need to have tampermonkey enabled for ACX, and your specific script turned on for ACX in the tampermonkey extension.
I don't know if Scott has banned public discussion of this, so I'm going to DM you on the specifics of implementing likes.
Thanks. The Pycea script looked easier, and worked fine.
Gave you one too
The past consensus has been not to enable likes, at least as a count display. I agree. I think they encourage group think and people tailoring comments to specifically attract likes, rather than discussion or information, clever insults and wicked burns ensue. Leave Reddit to Reddit.
Doesn’t the Substack app allow people to send a ‘silent’ like to a commenter? I don’t use the app myself but occasionally get a ‘like’ notification.
There is a tradeoff. On one hand, voting may encourage group think, in the bad sense of the word. On the other hand, voting allows people do get rid of spam, or users that were already banned but created a new account, without having to wait for a moderator.
That said, the Facebook-style voting, when there are upvotes but no downvotes, is probably the worst of both worlds, because you can't really use it to discourage spam and low-quality content.
I think Reddit with double votes has more of a hivemind than Facebook. I am kinda undecided about Facebook. Too many anti-science types, which is bad, but people with unpopular opinions flourish there, which is good. That is because hundreds of people can upvote a flat Earth stuff and no one can downvote it. In this case it is bad, but on the whole it is good to not have too much of a hivemind. Sometimes the accepted wisdom is wrong.
I was thinking about this a lot. Ultimately I would like a society with more diversity of opinion, even if it means putting up with cranks and fakenews and Russian propaganda bots.
Because now actual experts like Mearsheimer are not listened to if their opinion is different. Or Jeffrey Sachs. He was always a Russia expert, and he was always listened to in the White House, except during the Biden admin (and I think today with Trump, too)
Yeah, for nonsensical spammers there is the ‘block’ setting as the option last resort. Not ideal but sometimes necessary.
You will be amazed how much better your experience is when you block the worst 5% of posters!
I agree! I was reluctant to do it at first. I think it had something to do with that indignation dynamic that the Twitter algorithm exploited: There’s something about being infuriated by someone that kept me coming back from more. But once I blocked someone the indignant itch I’d been left with by their final post faded, and the silence was so sweet.
All that gets enabled by the script is a silent like. The person whose comment you liked is notified, but you cannot view how many likes anyone’s posts have gotten.
Okay.
https://citythatworks.substack.com/p/riders-deserve-to-feel-safe-on-the
This post came out too late to be part of the recent discussion on crime statistics and the perception of the same, but it would've been invaluable in providing vocabulary ("felt safety") and hard numbers on the criteria people use to evaluate their safety.
If there's ever a "Highlights from the comments on" thread, I hope I catch it in time to get this into the comments, because that's what most people ended up trying to talk about without really being able to articulate themselves or back up their anecdotes with data.
I read that post with interest and agree with your take on it.
Someone explain Iranian geography to me.
From the satellite image, Iran consists of a tiny strip of lush green on the Caspian coast, combined with a big swath of arid mountains which take up the other 95% of the country. So you'd expect everyone to live in the green part, right? Well no, as far as I can tell the green part is barely more densely populated than the rest of the country, and all the biggest cities are in the arid mountain part.
Now all the oil is in the south, so maybe that's part of the answer. But Persia has been the leading regional power for most of the past 2500 years. Apparently it has had 32 different capital cities over that time, and I'm not going to look up the locations of all of them but the ones I've heard of like Persepolis, Susa, and Ecbatana are all in the arid mountain bit.
How did a bunch of arid mountains come to support 100 million people and to spawn multiple regionally-dominant empires? And why didn't they all just move to the Caspian coast where it looks much nicer?
It's all path dependency, in an alternate Iran with a Latin American style leadership they'd have moved the capital from Tehran some time in the 20th Century.
The Caspian coast had severe malaria until the mid 20th century.
Look at a precipitation map and you'll notice a high precipitation strip (red in this picture) that runs diagonally from west to south.
https://commons.wikimedia.org/wiki/File:Iran_Precipitation_Map.png
Persepolis, Susa, and Ecbatana are in that strip or close to it.
I gave a similar question to ai a while back. I thought tehran was on the wrong side of the mountains and the gulf coast was grossly underdeveloped relative to its neighbors.
Basic gist is that tehran is a well defended mountain fortress located near strategic cross roads. From this location its power projects south, east, and westwards. Hence ancient empire stuff. It gets enough water off the mountains for mountain fortress purposes. But it struggles for major metropolis purposes. In modern times all power is concentrated in tehran so they build port infrastructure, but not cities along gulf, partially for defense reasons. They do have an important city in ahvaz a bit upriver from the coast. Expensive ground transportation costs are not immaterial for them.
Good questions. Is it insufferably hot and humid down by the coast?
If you look closely at a good population density map the Caspian coast is the most densely populated area outside of the Tehran metro.
Also, one of the things that we learned in Afghanistan is that a lot of people like herding sheep and goats. They weren't moved by promises of a more modern lifestyle. Success to them just meant more sheep.
Fog of War stuff. Unfortunately, so much has been going on, that I haven't bookmarked everything I should. Here are some interesting tidbits to consider that may or may not be true.
IRAN:
1. Is Mojtaba Khamenei really in charge now? He was the chosen successor to his father, and the Assembly of Experts was meeting to elect him Supreme Leader, thereby destroying the quorum required to make his selection legal. Supposedly, the surviving members met by phone the next day, and under IRG pressure, elected him. But rumors were that many clerics didn't like being strong-armed by the IRG, and they questioned whether he had the clerical credentials to be Supreme Leader. And there are constitutional questions about whether enough members of the Assembly of Experts remain to make the election valid. AFAIK, Western and Middle Eastern media seem to be hedging their bets and are wording it as Mojtaba Khamenei is "set to be chosen as Supreme Leader." Of course, the Israelis took out the offices Iranian TV network in Tehran, so if he were officially elected, the Assembly may have difficulty getting the announcement out.
2. The entire IRG senior command seems to have been taken out (according to Israeli sources). Regional missile commands seem to be functioning independently, which may be the reason we've seen missiles and drones being launched at every surrounding country except Afghanistan, forcing Gulf States, which would have preferred to stay on the sidelines, to react to the attacks.
3. But missile and drone attacks have been falling day by day. Quoting the X post below, "Clear downward trend: Peak saturation early (Day 1-2), then sharp drop as U.S. | Israeli strikes destroy missiles, TELs, storage depots, other launch infrastructure and crews."
https://x.com/UKikaski/status/2029300058017972422/photo/1
And there is some question about whether they have many more stockpiled, having sold the majority of their stock to Putin.
4. Why hasn't Trump pushed for Reza Pahlavi's return? He has a sizeable network of followers in Iran, and he's posted a sample constitution and reconstruction plan for Iran. Pahlavi would become a constitutional monarch and lead the transitional government under his plan. Does Pahlavi not have enough resources to bribe Trump to support him? And Trump claimed their preferred successor (unnamed) was killed in the attacks. Hmmm.
CHINA:
5. Western media and analysts haven't really absorbed the fact that Xi's removal of two of the three top leaders of the PLA had functionally decapitated the PLA military command. Generals Zhang Youxia & Liu Zhenli were removed (arrested? killed?) back in January. This only leaves a political apparatchik appointed by Xi at the top level of the PLA. Xi also purged their immediate subordinates, and State Media said that regional military leaders were instructed to remain on base and attend political training classes. This didn't happen, and various military commands announced maneuvers (with lots of videos being posted on X and other social media of troops marching in formation). Things have quieted down, but...
6. ...but the Political Bureau of the CPC hasn't endorsed Xi's autogolpe (top-down coup). They declined to officially remove Zhang Youxia & Liu Zhenli from their leadership roles at a Politburo meeting at the beginning of February and a second meeting last week. Under the operating framework of the CPC removal of disgraced officials from their positions is necessary before there can be formal trials. Zhang Youxia & Liu Zhenli are suspected to be dead. But there should be an announcement of support for Xi, and that hasn't happened.
7. Rumor is that the Xi faction has pissed off the Old Guard, plus a group of what they call the Reformers. Xi had already purged a bunch of Reformers from the PLA. Although Zhang Youxia & Liu Zhenli got their promotions from Xi, the Old Guard had come to regard them as their own. And notably, Zhang had been cautioning Xi that the PLA wasn't prepared to retake Taiwan.
8. And at last week's meeting of the Political Bureau of the CPC, they struck from the rolls a bunch of PLA regional command officers in the Southeast Region who were perceived as Xi supporters, and would have been the ones coordinating any invasion of Taiwan.
9. Around the same time, state media let it be known that Xi would be OK with the Taiwan status quo if the US didn't sell them advanced military systems. Also, Xi was supposed to host Trump in Beijing for a summit at the end of this month. He invited Trump soon after the initial meeting of the Political Bureau failed to endorse his autogolpe of generals Zhang & Liu, probably to boost his perception as a leader with capable international chops. Chinese and US representatives were supposed to meet in one of the Gulf States to work out the arrangements for Trump's State visit, but the Iran strikes and instability in the Gulf region have thrown those plans out the window. It will be interesting to see whether Xi continues to extend the invitation, because it could be perceived by Chinese hardliners as kowtowing to Trump after the Iran fiasco. And if he cancels, the Chinese perception is that American hardliners will perceive it as weakness on Xi's part. This is a no-win situation for Xi.
10. Reportedly, the Chinese were taken off guard by the massive strike on Iran. Their conventional wisdom was that the US would at most try for the nuclear processing center again. Planners weren't prepared for the US moves. Now they have to deal with the fact that Iran's oil isn't flowing to China; it could (will?) disrupt their economy. And the Iran fiasco has proved that Chinese weapons systems, which they sold to Iran, can't stand up to US weapons. The old guard & the reformers are very concerned. And Xi may not make it to the 21st Party Congress in 2027. Even is Xi wins his political battles, he won't be able to move against Taiwan in the foreseeable future.
RUSSIA/UKRAINE:
11. Russia has lost a major source of drones with Iran being taken offline. Under Ukrainian bombardments petroleum infrastructure in Western Russia, Russia's oil revenues have declined precipitously. Rumor (repeat, rumor!) has it that Putin is considering turning people's bank accounts into war bonds, and that he's considering universal conscription. Hardliners are calling Ukraine a distraction and urging him to go after the Baltics. Especially since they see the US is focused on Iran. It would take Putin months to call up 1 to 2 million more soldiers, and there are questions about whether they have enough weapons left in their stockpiles to arm them.
12. And Secretary of War Hegseth has reached out to Ukraine to assist with US drone defenses in the US Gulf (because our traditional missile defense systems don't seem to be working very well against Iranian drones). Zelenskyy responded favorably to the request, but he made it "clear Ukraine would help only on the condition that its own defence was not weakened and that there were diplomatic gains for Kyiv - suggesting, in particular, that Ukraine would be willing to swap its interceptor drones for more US Patriot air defences to protect against Russian ballistic missiles." (according to the BBC).
Well, we've been blessed to be living interesting times.
5-10) What you outline is good news for the West and bad news for China. Western analysts are aware (at least the smart ones, who are NOT talking to the media, being too busy with a war).
12) Ukraine claims their drones work, quite well. Now, they get a field test, so we'll see.
Last Wednesday was a big bombing day... (Remember, America wasn't ready for this one, they wanted a second aircraft carrier in the region, and Israel saw an opportunity and jumped the gun).
Old news: missile commander survived four bombings. Still alive? Who can say...
"and that he's considering universal conscription"
They simply don't have the equipment for that.
Not any kind of international relations expert, but when it comes to China, I am thinking of Napoleon's maxim not to interrupt your enemy when he is making a mistake. If the US is getting bogged down in a Middle Eastern war and using up all its best weapons and interceptors, just let it happen and keep your powder dry.
Great analysis, are you certain that China has supplied much in the way of missile defences to Iran. I thought Iran was supplying Russia.
Xi seems to be in trouble. China however is not just a cult of personality.
Yes. Although Iran has purchased a lot of Russian milware, Iran started buying from the Chinese, who offered a better price point. And the Chinese equipment seemed superior to the Russian equipment — it's an open secret that the Chinese copied US weapon systems whole-cloth. But there's corruption in China's military procurement systems. Maybe not as massive as Russia, but some analysts have pointed out that generals and officials have been cutting corners on QA and software updates and pocketing the difference. Pakistan also bought some Chinese weapon systems, but they failed miserably in a recent spat with India.
According to ChatGPT, Iran bought a bunch of China's HQ-9B long-range surface-to-air missile systems (and a bunch of other radar defense systems) to protect critical sites, but reportedly failed to effectively stop Israeli and U.S. strikes.
I'm curious when this is written about in 20 years whether the problem will be the missile defenses, or the men running them.
States in the gulf have a long history of buying top of the line technology and then failing to employ it effectively in live combat.
Possibly. But Ukraine has shown that RU missile defenses can be overwhelmed by drone swarms and drone misdirection. They can only fire some many expensive anti-missiles missiles before they can't respond any further and cheap drones overwhelm them. And UKR is using new cheaper interceptor drones for the slower stuff that comes over from RU.
It may be that the IRG are just missile defense dunces.I don't know how the US and Israel are taking out their setups, but I'm pretty sure it's not with drone swarms (because neither country has developed its drone warfare). But Pakistan had the same problem with same type of equipment Iran used, taking down Indian missiles when India struck some terrorist camps inside Pakistan.
Likewise, Iran is using relatively cheap Shahed drones to hit US targets and civilian targets in the Gulf, and the US hasn't been able to stop them all. So, I suspect that we'll be arguing about this with 20-20 hindsight for years to come. :-)
Both. The need for an "in person meeting" cost Iran dearly. Their dead-men switches were designed for an all-out assault, and that's not what the Saturday Massacre was. Hence the rest of the middle east now having a fig leaf to attack the heretical ummah (under the self-defense provision of Islam).
2. ”missiles and drones being launched at every surrounding country except Afghanistan”
Including Turkey? I read that Turkey had not been attacked, perhaps because of the ramifications of a direct attack on a NATO member. Is that old news?
4. “Why hasn't Trump pushed for Reza Pahlavi's return?”
It has been suggested that Trump’s reticence about long-term goals is intentional, to prevent such goals from turning into requirements. I think most of American, MAGA or not, has a pretty jaundiced view of “nation-building”.
More realistically, the goal is nation-destroying. Iran lashing out at the world in a final, desperate attempt for survival is only going to encourage everyone to finish the job.
Iran is literally surrounded by enemies and in that light, it's broad choice of targets has been fully understandable. Perhaps not the wisest choice, but an empathetic and understandable one. I don't think people fully realized before now just how deeply enmeshed Iran's neighbors are in America's broader military system, whether officially or de facto.
Of course, part of the reason why Iran has all these enemies is because they have insisted on exercising power through clients, and have refused to recognize the principle that one country shouldn't meddle in the internal affairs of another.
But however we got to this point, it's obvious now that Iran has an absolute strategic need for nuclear weapons. Despite what you may have heard, the nuclear program was subject to fierce debate in Iranian society. I am sure that is over now. I will bet you that every iranian, from the monarchists to the regime stalwarts, now recognizes that attaining nuclear weapons is the only safety for their country. If they had a consensus on this before, they probably would have built those weapons long ago.
As far as "destroying" Iran goes, that is a totally fanciful war goal. Even after two nuclear strikes, Japan was not destroyed as a nation. And there is no stomach in America to subject Iran to even that level of bombing. Or to allow Israel free reign to do that.
Read this again: Dead Man Switches.
The deadman switches were conceived with an "America goes full boar" mindset.
They weren't conceived with "surgical Israeli Strike" in mind.
So, the Iranians look foolish, and bellicose towards their neighbors -- the idea of their strikes was to get their neighbors to lean on the United States, but in practice, it's triggered their "self-defense" clause (no attack muslims without self-defense being involved).
Iranian nuclear program has been DOA for years upon years. Assassinating Americans is what got Iran in trouble, and nuclear weapons don't fix stupidity.
No, we will not "destroy Iran" as a country. There will still be an Iran, or something like it. Ukraine will disappear before Iran does.
Al Jazeera reported that as of March 4, 2026, Turkey's defense ministry confirmed that NATO air defenses intercepted and destroyed an Iranian ballistic missile (and in some reports, drones) headed toward Turkish airspace. No casualties or injuries were reported.
And for some reason, the Iranians lobbed some missiles at Azerbaijan. But AFAIK, they haven't lobbed any missiles at Pakistan since January of 2024. Not sure why. I thought relations with Pakistan were kinda tense.
Surely thats why. They attack countries that dont want to get involved and might pressure the US to stop. If they attacked pakistan theyd be inviting serious counter attack they are in no position to deal with
Despite supportive remarks from Azerbaijan to the Iranians at the opening of Operation Epic Fury, Iran attacked the main Azerbaijan airport to disrupt international air traffic that was now passing over Azerbaijan, which had been diverted from International air routes south of the Gulf. They also targeted an Azerbaijani elementary school. The drone missed the school, but you can hear kids shrieking in the video. Iran has denied attacking Azerbaijan, but President Aliyev has announced that Azerbaijan will respond militarily to the unprovoked attack. They've canceled all military leaves and announced a state of emergency. Don't know if their weaponry can reach far over the Iranian border, but their armed forces have combat experience (in their war against Armenia). It's worth noting that the northwest corner of Iran is populated by ethnic Azeris who carry a grudge about their treatment under the mullahs. An independence-minded Azeri group has just asked Azerbaijan for help against the central government.
Iran tried to hit the Georgian and Turkish links in the pipeline that goes from Turkmenistan under the Caspian Sea to Baku, down to Turkey, from which Israel gets a lot of its petroleum and gas. I guess they missed, and it's still running.
https://www.euronews.com/2026/03/05/aliyev-vows-attacks-on-azerbaijan-will-face-our-iron-fist-after-iran-drone-strike
No one knows why they launched an attack on Cyprus. There are no US military bases there.
Dead-men switches don't respond to "current intelligence." they respond to projected future intelligence, based on when the plans were drawn up.
(And someone may have wanted to spoil some Israelis vacations in Cyprus. When you're to the point of using dead-men switches (aka preprogrammed orders), the artillery officer gets to choose. "Yes, No, Point Somewhere Else"
Interesting, thanks.
Thank you for this summary of world events. I could’ve spared myself hours of doom scrolling if I had come here first.
The average commenter here seems deeply jingoistic and in favor of an aggressive foreign policy against any country perceived as an enemy of the West. With their very technical interests, I imagine the average commenter doesn't have much time to study these things, and so when it comes to foreign affairs, they are left to passively sign off on the manic aggression of mainstream corporate media analysis. This is the last place I would come to learn about world events and foreign policy.
The OP didn’t strike me as terribly jingoistic. I thought the scuttlebut about what is going on in China was interesting.
I would argue that anyone who isn't jaundiced and cynical about this Iran operation is regrettably jingoistic and militaristic.
Hell, anyone who defaults to a neutral, "let's wait and see" attitude about American military operations is passively jingoistic, and is likely naive about America's post-1945 military legacy.
Are you in favor of propping Xi up with military action? Do you see this as an aggressive foreign policy, against a state that believes Trump is an existential threat?
America's foreign policy "ideology" is quite a bit different than you'd think. Obama put it best, "America doesn't have allies, only interests."
If it's in our interest to prop up Xi, we will. If it's in our interest to assassinate Zelensky, we will. Etc. America goes around "color revolutioning" many countries, and I'm sure you cheered for the Arab Spring, which was a very aggressive foreign policy decision that has paid off extraordinarily poorly.
I consider Obama to have been insanely aggressive when it came to foreign policy and military affairs. Make of that what you will.
Then I'm going to count you as one of the few principled...isolationists? Something like that*. Few people can stomach "Obama was cuckoo-crazy too" (let alone Biden, who had former military Democrats saying "he's trying to start World War III" -- scuttlebutt supports this, alongside his appointment of neocons to run Ukrainian policy.)
*wanna call yourself a "chary interventionist"? or just "not a chicken-hawk"?
As a confirmed peacenik, our Obama/Clinton/Biden-era Ukrainian policy has me pretty ticked off.
I'm also a peacenik. I only believe in using military force as a last resort. Since 1945, I can't think of a single American military intervention that was honorable or necessary. Perhaps there was some obscure operation that was worthwhile, but certainly none of the headline grabbing operations were reasonable.
Funny, I read hella lot of stuff every day. Happy to do the doom scrolling for you. For whatever reason, it doesn't upset me as it does my fellow Lefties.
And as long as Trump doesn't put US boots on the ground, I'm OK with Operation Epstein Fury. Fuck the Mullahs. They didn't deal with Obama honestly when he tried to get them to scale back their nuclear ambitions. Best to take them out while we still can.
Bibi's been saying the Iranians are "very close to a nuclear bomb" since the early 1990s. "Scale back their nuclear ambitions"? I don't think they HAVE nuclear ambitions, to be frank and honest (there's good recorded evidence from around Bush's Axis of Evil that the Iranians didn't want a bomb if it would mean an invasion). I'm not sure our military assessment says so either.
That said, putting assassins on American soil hunting Trump and other military was a very bad ploy, and worth toppling their government over.
I am sympathetic to that position
Quick note on point 11: Russian Shahed production is almost entirely domestic, with a primarily Chinese (not Iranian) supply chain. I don’t expect the war with Iran to have any significant impact on Russian drone production.
My understanding was that many of the electronic and mechanical components were being imported from Iran. Also, Iran had sanctions-avoidance network in place for money laundering and electronics imports, so RU has probably lost this for at least now. And now that UKR has taken out two Sahed manufacturing facilities with their flamingos, we'll see if RU can keep up their manufacturing pace. Also, a bunch of electronics manufacturing facilities were taken out pre-Flamingo. I don't know if RU has gotten them completely back online. But RU is still getting German electronics for their radar and anti-aircraft systems. And China is providing electronics for RU's smaller non-Sahed drones. There was a good write-up on this last year. Sorry, I didn't bookmark the site.
My understanding is that the original licensed production run of domestic Shaheds was essentially from Iranian kits, but since then Russia has modified the design into a number of different variants that haven’t been shared with Iran. They’re definitely using a significant quantity of western-sourced parts, although my understanding is the majority (maybe just a plurality?) of them are coming through China and few if any are coming through Iran.
What you say is likely true. However, Chinese electronic components, including Western-branded semiconductors, reach Russia primarily via Hong Kong middlemen, then on to Iran, Turkey, and some Central Asian countries. Chinese manufacturers want plausible deniability that they're not avoiding sanctions. HK is a big transhipment point for US technology into sanctioned countries.
Regarding 4: There is a big difference between the Iranian population being upset at their government, and the Iranian population wanting their government to be replaced by an American puppet. You’ve been reading too much CIA propaganda if you think Pahlavi is viable barring a full invasion.
All those protestors were waving the old regime's Lion and Sun flag. Seemed like there was a lot of popular sentiment for the Pahlavi on the ground. And your understanding of the US deep state is faulty. The CIA doesn't have a propaganda arm anymore. The State Department does, but since Trump zeroed out their foreign-language communications budget, they're not putting out much propaganda material anymore. As for internal US propaganda, that's handled by Fox News.
As for putting U.S. boots on the ground in Iran, I'm sure that would drive popular opinion the other way. And let's face it, the Pentagon can stir up the hornet's nest, but they don't know how to deal with asymmetric warfare (and one could argue that they haven't fared well in any long-term ground war since WWII). Let Pahlavi return. It's the best realpolitik solution.
The thing those protesters were conspicuously *not* waving, is likely to be decisive. They were not waving guns.
Almost everyone in Iran with an actual gun, is solidly on Team No Pahlavis. If they weren't, they would not have been allowed to have guns. OK, there's probably a some people in the regular army who'd be OK with Pahlavi and whose technical skills made them un-purgeable by the IRGC, but not enough to matter. And there are some armed Kurdish insurgents, but they're doing their own thing, and there are maybe some resistance groups that stole some police guns a month or so back, but again not enough to matter.
What you're seeing under the Lion and Sun flag was the minority of Iranians who are largely urban, educated, English-speaking, and unarmed. What they want doesn't matter now, because all the people with guns are looking for *someone* to shoot and they can't shoot Americans or Israelis because an AK-47 can't reach 30,000 feet. They're going to be keeping their heads down, or they're going to be dead.
If you want a Pahlavi regime, that's going to take American boots on the ground. Either a *lot* of them, or a lesser number of special-forces advisors working to build up an effective Iranian resistance over many years. I'm pretty sure Trump isn't up for either of those.
According to some news reports, the IRG finally had to bring in a few thousand paramilitary groups from their network of sympathetic terrorist groups to put down the protestors. During the protests (and before Operation Epstein Fury kicked off) there were explosions at IRGC facilities and bases, fires at missile depots, attacks on military warehouses, and sabotage of communications or power systems. Khomeini (the elder) went into hiding, and the IRG was kicked back on its heels until they received reinforcements from their external assets to put down the insurrection. Some have blamed Mossad for the attacks, but Israel has remained silent on them. And some have suggested that the attacks were carried out by Mossad-trained Iranian dissidents. Israeli intelligence has been working with impunity in Iran for a while. Last year, Mahmoud Ahmadinejad admitted that an Iranian counter-intelligence team tasked with locating Israeli spy cells was led by a general who turned out to be working for the Israelis. And someone tipped off the Israelis precisely when and where Khomeini convened a high-level meeting of his cabinet and his advisors, decapitating the senior levels of government.
The protests in Tehran kicked off after a major scandal in which Tehran's water supply was diverted to agribusinesses controlled by senior government officials. Tehran was running out of water, and Iranian President Masoud Pezeshkian stated that moving the capital from Tehran to the south was "no longer optional" and is essential due to severe water shortages, over-expansion, and land subsidence. But it was clear the agribusinesses were getting all the water. Moreover, government contractors were charging exorbitant prices for the water they trucked into Tehran. This, on top of rampant inflation and job losses, pushed Iranians out into the streets. I suspect that most would be happy to get rid of the mullahs.
Israel's responsible for some part of the water shortage, due to damming up rivers upstream in Afghanistan. I'm willing to call vanishing the water supply a war crime (as I've done when Ukraine did it to Crimea) -- it's punishing civilians.
Protests are one thing (Iran is often protesting), but explosions et alia sound an awful lot like CIA/Mossad (maybe our own milint if you really want to say CIA is out of the game -- which I maintain they're not).
The whole 21th century Middle Eastern experience seems to indicate that local "puppets" have a lot more agency that Americans think. This is a region where political scheming was perfected long before Columbus sailed the Western ocean, and I absolutely wouldn't be sure of who is a de-facto client of whom.
Or 20th century experience too. Saddam Hussein in Iraq, the Taliban in Afghanistan were supposed to be anti-Communist "American puppets".
Indeed the entire Islamism, which caused a lot of terrorist trouble in the West, seems largely a Western creation. Back then Brits encouraged Wahhabism, Salafism, as the Wahhabites really hated the corrupted and nearly atheist Turks. Then America encouraged them further because they really hated the atheist Communists. The Taliban got a massive number of Stingers when they were fighting the Soviets. That really mattered, because as they said "we are not afraid of the Russians, but we are afraid of their helicopters".
I heard that the rural parts of the country are more religious and much less West leaning than the population in the cities. But I don't know how to check that.
That's the way it is everywhere else. It would be odd if Iran were different.
Imagine how skewed your perception of the American electorate would be if your only first-hand information came from protest videos from big cities.
I think that would make sense.
I think that it's a combo of:
- The "have someone more pliant from inside the regime replace the old guy" seems to have worked out for Trump in Venezuela, so he's trying to repeat the model (for one thing it's very much not the Iraq 2003 style total upheaval nation-building project, which Trump admin is very insistent it's *not* trying to repeate)
- related to the above, Reza Pahlavi just plain gives rather strong Ahmed Chalabi vibes
- this whole "combat operation" still feels like they're flying by the seat of their pants, which is of course Trump's general MO for most things (it's got him this far!), and strong commitment to *any* option would not facilitate this model of operations
Trump seems to be going for "let's bomb the hardliners and leave the moderates alone" with a strategy that consists of: "continue to bomb until someone starts crying Peace!"
Venezuela was a specific "take Venezuela free oil off the board for Cuba." He didn't need much buy-in from the next Venezeulan dictator (and the "we own you, wear what we say" was kinda explicit).
Moreover, if Exxon/Shell/Chevron wants its oil facilities back online, they're going to have to spend tens or hundreds of billions of dollars to rebuild the facilities that rotted under the Chavenista's neglect. I'm sure Delcy Rodríguez would settle for a smaller piece of a much bigger pie.
This is Trump's specialty: crafting win win win deals for everyone. See the Abraham Accords, which is really a pipeline deal.
My closest Persian friend, an American citizen for at least two decades but with family in Iran, was disappointed that the Fordow attack last year was so limited. “Why did they stop?” he asked me.
But he is disappointed by the “cut off the head” strategy, which struck *me* as brilliant and measured, like Israel’s pager attack. He says the “Muslim Mafia” is too big and deep for this to work and fears we will eventually make peace with some low-level Mullah, every bit as fanatical as Khamenei but perhaps more deceptive, and nothing will have been accomplished.
In the interest of harmony I didn’t ask him what he thought we should do. I understand that what’s good for America and what’s good for Iran may not be identical.
One guy was invited to a Persian restaurant by an Iranian friend, then the door opened and 8 incredibly arrogant guys walked in and the Iranian friend started cussing them out. They were the consulate people and everybody around the tables hated them. The point is a real revolution would mean a massacre of elites and they know it.
> He says the “Muslim Mafia” is too big and deep for this to work
Yeah. Not sure how valid is this analogy, but it would be kinda like killing Hitler, and then letting NSDAP choose the successor.
After the Fascist Grand Council fired Mussolini, they asked the king to find a successor, who found Badoglio, a largely non-political soldier who installed a government of bureaucrats and soldiers and the Fascist Party was disbanded. He had an impossible job, to end the war without being invaded by Germany. So he tried secret negotiations and it did not work out too well.
Isn't that pretty close to what happened? Hitler killed himself, and the successor he had chosen, Karl Dönitz, soon ended the war by surrendering.
That only happened because we first completely destroyed Germany's armed forces and industries, conquered and occupied most of Germany and made it clear that we were taking the rest in a matter of weeks at most, and killed several million Germans in the process. Absent that, Hitler wouldn't have killed himself and if he had been killed, whoever succeeded him would not have surrendered.
Even the Stauffenberg variant of Operation Valkyrie would not have resulted in (immediate) unconditional surrender.
The United States may have mostly destroyed Iran's Navy and Air Force, but it is exceedingly unlikely that it will destroy the Iranian Army (or the IRGC's land forces). There has been very little in the way of industrial destruction that I know of, and a policy of wholesale deindustrialization seems also unlikely. The idea of American armored divisions marching on Tehran seems frankly absurd. And while it's possible Trump is willing to kill several million Iranians, that would be difficult and tedious to do without nuclear weapons.
So, yeah, meet the new Ayatollah, same as the old Ayatollah. Even if we play multiple iterations of that game, we'll keep getting the same outcome.
My Iranian friends were frustrated that Trump didn't hit them while the protests were in full swing.
You need to be wary about Iranian friends in the west. They have fled the regime and are antagonistic to it, and are often antagonistic to the entire country. Like White Russians who were extremely anti communist and often anti Russian. It’s probably not that popular in Iran to support either Israel or the US, I’m sure the bombing the children’s school and the recent mass killings in Palestine don’t help either.
Iran is actually a republic. They have a democratic process. It’s flawed but it’s there. Replacing the top guy is all very well but it’s still replacing a dictator (of sorts) with another. That happened before and it didn’t end well. Since there are elections and a president it seems to me it’s just a matter of replacing the role of the supreme leader. Is that being discussed or is the Trump regime relying on people’s ignorance of the internals of Iran to convince us that he has a guy.
Theocratic democracy, yes, just like Israel (whose judiciary is the compromised part).
I can certainly sympathize with that. I presume the reason was that they were waiting for solid intel about the whereabouts of the first targets. It’s unfortunate that the timing worked out so badly.
That's what Israel was waiting for. America was waiting for the second carrier, which notably didn't get there before Israel struck.
Further, perhaps more empirical arguments against superintelligence. The Transformer model, the basis of all modern AI, was only intended for machine translation. However they trained it on texts that contain reasoning, and it picked up that pattern. This is delightful so it now gets trained on a lot of philosophy and science texts. But the upper limits of its reasoning are these texts. So 140IQ.
Of course the 140 IQ combined with vast knowledge, very fast thinking and the ability to communicate with millions of people can STILL be totally dangerous. People will be tempted to give it too much power in order to beat the competition or win a war.
So I am not arguing against AI could make humans extinct. Just give it nuke launch codes aka SkyNet and we could all die. Or killbots + automated killbot factories. I am arguing this does not even require superintelligence.
So the Terminator movies were directionally correct. SkyNet was mega powerful because of all these, and nearly made humans extinct. But it was not a superintelligence, humans could still outsmart it frequently enough.
Although I will also say that one thing was not correct, they expected deterministic algorithms, so SkyNet was understood as too rigid, not as creative as humans. This is not so with AI which can get very "random" thus creative depending on where the temperature slider is set.
I don't think you get to 140 IQ with an unthinking LLM unless you can curate the texts and say "these are the ones by the geniuses; weight them much heavier than the rest". And the volume of text required, I think makes that impractical.
The current approach seems likely to asymptotically approach the intelligence of the average internet user, albeit one who has recently done all the reading in the subject at hand. And that's not to be underestimated; cheap fast well-read mediocrity can change a lot of things. But I agree that it's not a path to superintelligence, without adding something currently unknown.
You could also do stuff like ‘train it on the arxiv’ which will asymptote it to the average intelligence of an arxiv poster (probably higher than an average internet poster), albeit in a narrow domain.
I think current model Chat and Claude, when analyzing physics and well prompted, are pretty close to a mediocre professional physicist who has recently read the entire arxiv.
Within my narrow domains of expertise I could give you better instant answers than AI, but step a little outside those boundaries and I could not. Well probably I could but it would take me 5 months instead of 5 seconds.
Yeah, I can see Arxiv having enough good semi-curated material to constitute a reasonable training dataset within its domain.
Well, at least until the LLMs start publishing on Arxiv, then all bets are off. Fall back on peer-reviewed journals, and watch the entertaining legal brouhaha of OpenAI vs Elsevier?
I think that's true for classic LLMs, but reinforcement learning like we're seeing now can easily surpass human ability in selected domains.
Only in domains where you can get ridiculously large amounts of training data with success and failure being graded automatically and cheaply. If you want say a chess-playing LLM, sure, you can have it play bignum chess games against different instances of itself. The ELO scores speak for themselves, and the math is simple. If you want instead an LLM that can fly an airplane, OK, you can throw bignum instances into flight simulators, but that's a lot more compute and it's going to leave blind spots in those aspects of aviation that aren't well captured by flight simulators.
Most of the interesting problems are going to be in the second category, I think.
IQ isn't conserved via textbook. If it were then it would be impossible to produce a human who was smarter than the smartest previous author. What book taught Einstein to have an IQ of 180?
Einstein never had his IQ measured. That's just a bullshit estimate by Catherine Cox, an American psychologist, and one of Lewis Terman's team studying gifted people. In 1926, she published a study titled _The Early Mental Traits of Three Hundred Geniuses_ which estimated the IQs of famous doers and thinkers. Goethe beat Einstein's estimated IQ with something like a 220 IIRC (which isn't a valid number by today's IQ tests). Cox just made estimates that Einstein had a super IQ because none of the ~1500 geniuses that Terman tracked throughout their lives did much of anything noteworthy.
Einstein himself was modest about his intelligence, saying, "I have no special talents. I am only passionately curious". He further emphasized persistence over raw talent, saying, "It's not that I'm so smart, it's just that I stay with problems longer". I believe he also said that his first wife, Mileva Marić, a mathematician, was smarter than he was, but I can't find any quotes, so I may be misremembering it.
I have no idea what Einstein's IQ was - that was just a random above-average guess. The point is that he was far more intelligent than any of the authors of the textbooks that he read as a student. The best scientists rarely write widely-used texts.
Modesty is poor evidence. All it measures is personality factors. Most famous scientists say that their intelligence is nothing special. The only thing that that is evidence for is that most scientists care about their social status. It's beyond doubt that their intelligence is, in fact, far above average.
>I believe he also said that his first wife, Mileva Marić, a mathematician, was smarter than he was
Yes, men frequently lie about their wives because it makes their personal lives much more comfortable. There is zero evidence that Mileva ever had a scientifically important idea. They separated in 1914. Albert published many important papers after that point. Mileva never published anything and there is no evidence that she ever did anything privately. She never even finished her degree because she wasn't able to pass the final exams (she failed twice). The notion that she contributed in any way to relativity is nothing but revisionist feminist nonsense. She was nothing but a housewife.
Most "famous scientists" KNOW their intelligence is nothing special. That's part of what makes a genius, testing small ideas and fitting them together. Geniuses and "sub-normal intelligence" folks have a lot in common. It's the midwits, who memorize things, that think "they must be Very Smart and Very Special."
Geniuses don't need to be "extraordinary intellects" -- they just need to work harder, more persistently, and develop quality control. Intelligence is something that you can continue to build your entire life.
No, they know that intelligence by itself is useless. Hard work is also essential. Since intelligence is out of their control but hard work isn't, they adopt the attitude that "intelligent isn't important" in order to motivate themselves to do hard work. They're wise enough to know that if they sat around thinking of themselves as brilliant that they'd never do anything.
Great scientists have both extraordinary IQ and extraordinary work ethic. No one of normal intelligence has ever won a STEM Nobel.
"No one of normal intelligence has ever won a STEM Nobel." https://en.wikipedia.org/wiki/Philipp_Lenard was an experimentalist who had almost no understanding of theoretical physics. He was just a relentless experiment-builder.
Interestingly, Nazis (like Lenard) disliked IQ testing, because they thought it measures "Jewish tricks" and they attributed success to hard work, not IQ. Lenard was apparenlty a hard worker and not a good thinker. Edison also attributed his success to work, not IQ, he had no theory of lightbulbs at all.
I don't really know who is right in this. Just wanted to show this perspective. Success through "brute force" is a very typically German thing.
Intelligence is hardly out of a genius' control. If it were, they would have continued to test as an idiot on the IQ tests*. Increasing one's intelligence is done through hard work and better models. And as you reach diminishing returns, you STILL don't feel intelligent -- because one of the fruits of true intelligence is capturing the uncertainty of your knowledge.
*People aren't born intelligent. They grow that way, and it's a painstaking process that children who merely mirror the teacher never truly experience. Hence why "gifted" children so often have middling careers.
I agree about Einstein's wife, but with regards to Einstein's IQ, intelligence is not the only important thing, there are many other factors involved, including pure luck. Therefore, if you choose e.g. 1000 people with IQ above some threshold, the most successful one of them does not necessarily have to be that smartest one, or even one of the smartest ones.
I mean, Einstein definitely *was* super smart. I only object against attempts to estimate it by comparing to achievements of other people. More precisely, high IQ is necessary to invent (or even understand) some things, but is not a guarantee of success, even if you try to do something like that (i.e. ignoring the fact that different people specialize on different things). I can imagine a parallel universe where Einstein had a different boss, or got sick, or had too many problems at home and couldn't focus on research, etc., and in that universe he perhaps didn't invent relativity, but his IQ was the same.
Oh of course. High IQ is a necessary but not sufficient ingredient. The world is full of lazy geniuses who never did anything.
Are human-generated texts limited to human IQ?
To some degree, we can substitute time, money and headcount for raw intelligence. If 10x130 IQ researchers can write a more useful textbook than 1x140 IQ one, it follows that 10x140 might be able to outperform 1x150. I could believe that this gives diminishing returns, but it doesn’t obviously prevent the creation of higher-IQ text.
Yeah there is a pretty standard ensemble argument from classical machine learning that suggests that a "panel of experts" that pools information can outperform any individual expert. You can achieve this even with simple models, e.g. polynomial regression + random forest + simple one-layer neural net. It does not give you unlimited gains, but seems pretty reasonable that 3 x 150 IQ models gives you >150 IQ. And your "time tradeoff" idea seems right as well - in a lot of ways the difference between a 150 IQ and 120 IQ is moreso "processor speed" - a 120 IQ person is still capable of learning multivariate calculus, it just takes longer. And many of the concerns about powerful AI still apply even if you "merely" have Dario's country of geniuses in a datacenter (100 million 150 IQ people at 10-100x human thinking speed) vs. the Machine God with IQ = 10,000.
I don't think handwaving is quite enough to support the logic. You set out some facts and give some conclusions, but I think extra work needs to be done to show that the conclusions follow from the facts. Specifically, humans are clearly capable of writing down ideas that were never in their training set. Can you articulate why you believe the set of written material ingested imposes some kind of constraint on computation performed in silicon that is not imposed on computation performed in meat?
Personally, I agree that modern AI is limited, however I suggest that the limit is more likely to arise from our choice of training model / loss function than from the transformer model. A network of the latter can be used to build aribtrary systems. It is the combination of the former with the training data that determines what systems actually get built; and I don't believe that the missing piece is the training data because that's all /humans/ have to work with and it doesn't stop us inventing things.
Because the Transformer was made for machine translation, and then it just randomly picked up reasoning patterns from the training material.
So, what I read in your sentence is that instead of being constrained to do the task it was originally designed for, the software became able to do things its creators did not design or anticipate. This seems to be an argument for the exact opposite of the conclusions you are trying to use it to support.
If you intended my takeaway to be something else, please could you expand on what that is and how it follows from what you've said?
Well my point is that it can horizontally deviate from design, that is, do something else that was designed, but I think vertically not, because if it is not in the training data then not able to. Others pointed our reinforcement learning, yes, if there is a clearly defined goal, such as winning chess, and there is self-training possible, like simulate two strategies against each other. But when we are talking about superintelligence, we generally mean something that can talk humans into giving it control over nukes. Superintelligence - I am not moving goalposts, this is literally why Yudkowsky considers it super dangerous - is primarily about an extremely good communicator who can talk humans into giving it power.
When "the training data" is literally the entire text output of humanity, and this is sufficient for humans to come up with things that humans before them did not and could not, I am not really sure I understand what it is that is "not in the training data" that could be.
Have you considered the effect of the choice of loss function on the result at all? I've been talking about that this whole time - about the fact that it is the loss function that determines what it is we use the training data to actually build - but you've not mentioned it in any responses, so I am left wondering whether there is some reason you are focusing on the training data instead.
>Can you articulate why you believe the set of written material ingested imposes some kind of constraint on computation performed in silicon that is not imposed on computation performed in meat?
Just to hazard a guess here. The only thing that occurred to me is highly random associations that turn out to be useful. And are largely informed by our intimate participation in meat space.
I get the first sentence. The second seems doubtful.
Well, I agree with you in that it remains to be seen, but that’s the way I lean.
I think the physical process of living in the world affects the way we think a lot more than we credit.
I don't have any argument with that last claim. I just don't see any reason to suppose that it's a precondition for thinking in general.
Its not, I agree.
Reinforcement learning can let AI pass human IQ and reasoning capabilities, see chess or go for example.
Exactly. In these cases the AI could generate training data, indefinitely, by playing. If we were able to give LLMs a similar environment, maybe we could do the same. I'm sure many labs are working on that.
Google did this with other non LLMs, see deep mind blog on ai agents playing various 3d games.
Ilya on Dwarkesh podcast discusses the difficulties with training on reasoning chains that get too long, however.
Assuming that the constraints of human IQ tests apply to LLMs is nonsensical, for too many reasons than I'm willing to list here. I'm not saying LLMs aren't limited by the intelligence of their training material, which is obviously true. But trying to identify that limit via IQ is...like trying to measure the energy output of an atomic bomb based on the weight of the entire bomb and nothing else. Not a perfect metaphor but whatever. The real answer, for this whole spectrum of questions, is "we don't know."
Edit: a better metaphor might be: building a two-legged humanoid robot and estimating its top speed based on fast humans.
I agree with you that applying IQ to LLMs in this way is somewhat silly; however, I disagree with you when you say that "the real answer ... is 'we don't know'". I think it's obvious that LLMs cannot become "superintelligent" (to the extent that term has any meaning) or even as intelligent as humans (ditto), due to the way their architecture works. Unlike humans, LLMs are e.g. unable to learn on the fly (re-training is incredibly expensive), to fail gracefully when confronted with novel inputs (or occasionally even ordinary inputs), or to synthesize many different inputs (other than text). Some of these weaknesses can be mitigated somewhat (usually by dramatically increasing computing power), but ultimately the problem stems from the underlying NN architecture.
Granted, it should be possible to use LLMs as a component in some other AI architecture to produce an intelligence that thinks more like humans do, but thus far no one had managed to do it, and it doesn't look like it will happen anytime soon.
That said though, a tool does not need to think like a human in order to be useful, and indeed quite powerful: for example, spreadsheets don't think like humans at all, but possess superhuman powers of calculation. And a tool does not need to be superintelligent in order to be dangerous in the wrong hands; for example, bulldozers are really stupid, but can cause a lot of destruction when improperly applied. The more powerful the tool, the greater the danger, and sadly LLMs are quite powerful...
Just in-time for the End of flu season, I came across a podcast from last fall that included a segment on getting your flu vaccine—I've never had issues with mine.
One of the things they say about Influenza is that it can make one susceptible to Bacterial Pneumonia. Isn't that Also a possible escalation for a cold, bronchiolitis, and a crapton of upper respiratory infections that Aren't the flu?
Yes. The teaching in medical school is that are lungs are constantly exposed to germs entering from our mouths (think swallowing down the wrong tube) and from our environment. Inflamed or damaged lung tissue is both a more favorable environment for the growth of bacteria and also less able to fight off the infection. Influenza, as compared to other viruses, is more often associated with severe presentations that are likely to predispose patients to bacterial infections.
Here is a source with more information. https://pmc.ncbi.nlm.nih.gov/articles/PMC5324726/
Thanks!
I especially like the list of how certain infections progress, with pneumonia in Every Single Category.
Isn't Anthropic, OpenAI and Gemini already a mass surveillance device?
No.
Zvi covers it somewhere in here: https://thezvi.substack.com/p/secretary-of-war-tweets-that-anthropic?
> As mentioned above, the US government doesn’t have a formal legal definition of domestic mass surveillance, only “bulk collection.” And the US government has, basically long maintained that even if they hoover up a bunch of information indiscriminately, they haven’t done bulk collection so long as their individual queries against that mass database are more targeted when humans look at them. As a result, at least one Director of National Intelligence said under oath “no” when asked “Does the NSA collect any type of data at all on millions or hundreds of millions of Americans?” even though NSA has admitted it does by the ordinary meaning of this question.
In other words, if you choose to look for something, you will certainly find it
I am very familiar with USGOVT system and know several of the TIA architects. And efforts before TIA (and even back a bit to Operation Ivy Bells). US Govt is very good at say node analysis in a time series to know what's going on, despite the package being encrypted. Walz got caught up in that way with his Signals-encrypted message recently. No warrant required, but useful for influence.
My point has nothing to do with USGOVT. Anthropic, through user prompting, has the equivalent of mass surveillance. They know your concerns, tone, typing speed, voice, etc. They could gain huge knowledge about you, and even use Claude to steer you, and are not subject to the rules put on the USGOVT. Same is true for OpenAI and Gemini. They are unregulated in this realm. Other than market forces.
I can most certainly say the USGOVT has thought about the various ways this can go wrong much longer and deeper than Anthropic, OpenAI, and Gemini.
Anthropic might have a lot of user data, but they aren't buying cellphone tracking data the way the government is. They can't tell that you went to a protest unless you deliberately inform Claude of that fact. The government can.
Anyone can buy cell phone data. Is that what people think is mass surveillance? Anthropic type platform is a non government version of Telescreen. The technology could easily become a vehicle of thought police, whether they are private or USGOVT.
PS: USGOVT can track the movement of every cell phone around the world without buying the data. Likely has been able to do that since cell phones were created. They don't need anthropic to do that or analyze it. Anthropic would have helped when Tia was built, but that is old tech. USGOVT buy data for Parallel Construction.
Go to a motel, and get your license plate captured (this happens every day, to every motel in America). Mass Surveillance. (This is a Private Eye network of folks, not governmental).
No, the government has to get a warrant to access that data which is only granted for specific individuals
The US needs to obtain a warrant if it seeks to prosecute. They don't need to do it for a foreign person. But that is not my point.
Anthropic has been worried about USGOVT. But Anthropoc can, synthetically, gather a lot of invasive information through your prompting. So they are de facto a mass surveillance data. They also have the means to steer you in their response. Likely a far more serious violation of individual rights than how NSA operates CONUS.
People miss that you can synthetically replicate private data from more public data.
Is it stored, and then a warrant lets you look at what is stored? or do you need a warrant proactively to get anything recorded at all?
If the former, it's still a mass surveillance device.
At least with ChatGPT, Users can choose to have off-the-record conversations which are not stored / trained on by default. I presume a court would need to provide a wiretapping warrant to obtain those. As ever, it remains bad opsec to produce electronic records detailing illegal acts.
Transformers are not the path to ASI or even AGI. A transformer which has the global optimal parameters to minimize the loss for the standard internet corpus out of the box, is not going to be AGI
Transformers have nothing to do with what training objective function or data is used.
Even if it did, LLMs are also not minimising the loss of internet corpuses since ~GPT-3 era. We have _at least_ 2 generations of breakthroughs in that time (RLHF + RL on CoT).
Why not?
Because human beings are still better at navigating the semantic ocean. seriously. I have come across this with Claude on many occasions and we have discussed it. I posted it a text by Franz Kafka once. I asked what it thought the most striking thing about the story was. It came up with a very concrete thing that was observable; something to do with the transition from the realm of thought to the realm of physicality. I came up with a different one. The one I came up with was knitted semantically across many domains that we had discussed in detail and embraced things that it knows about our previous discussions. (I say this so certainly because in other conversations, it has referenced those things so it’s clearly included in it’s general memory of dealing with me.)
I doubt those sorts of things can arise in a second order derivative of living-in-the-world.
Will there ever be an AI that can pass a Turing test of convincing you that you are in a conversation with Robin Williams?
I just realized superintelligent AI is not possible and it had to be obvious all along. First of all, can we say the model is intelligence and training is knowledge? So the model itself is a traditional software that is designed traditionally, right? So the question is, how could you design something smarter than yourself? Neither a human nor self-coding AI can do that because you just cannot imagine what it is like, what it is does. Can a 8 years old design an adult? Yes, he can give us the specs of his ideal dad. But we adults would not necessarily agree with that spec. Besides, there are other aspects of being adult. The 8 years old cannot design the thinking process of a 30 years old programmer. Self-coding AI cannot do that design either.
Or you can argue that intelligence is not only in the model design, but in the training too. In that case just where can you find examples of superhuman knowledge to train it on?
On the other hand, AI does not need to be superhuman to be really powerful. 140 IQ with immense amounts of knowledge and very fast thinking speed and ability to communicate with millions of people simulateously would be very powerful.
> So the model itself is a traditional software that is designed traditionally, right?
No.
Humans can absolutely design a smarter than yourself person. Autists do it all the time, because they don't start with the basic inbuilts. When you get used to building, say, facial recognition yourself (or emotional-reading of faces), you can learn to build a lot into your brain.
> First of all, can we say the model is intelligence and training is knowledge?
No, we really can't. You can't separate "intelligence" and "knowledge" in LLMs, they're fundamentally entwined. This is decidedly not like the separation between, for example, the ALU (processing) and the RAM (storage) in a CPU. Any argument you're trying to build on this assumption will fall apart.
" You can't separate "intelligence" and "knowledge" in LLMs"
Nor in humans, for that matter.
By this logic, it is also impossible for a human to physically assemble a machine (eg. a hydraulic press) that is physically stronger than that human.
Also, please note that that 8-year-old will eventually become an adult, despite his inability to "design" an adult.
Here's a hypothetical to consider: intelligence *may* be primarily a function of available computing power. Given what we now know, it is reasonable to suppose this may the case. *If* that turns out to be the case, then a person with IQ of (for example) 140 can easily design a machine with a much higher IQ, if they know the basic algorithm involved in turning compute into intelligence. All they have to do is implement the algorithm on a machine, then give the machine more computing power than their own brain has. All that takes is to make the machine bigger.
Yes an 8 year old can design an adult.
Here's the spec: 'Have a baby, raise them for 18 years'.
You don't have to understand everything a process does or produces in order to design the process.
I've programmed genetic algorithms that mutate parameters of their code each iteration until they converge on certain target metrics, and I had no idea what solution they would end up with at the start of the process.
Generative designs are old news at this point. We've been growing things without understanding them for decades.
> So the question is, how could you design something smarter than yourself? Neither a human nor self-coding AI can do that because you just cannot imagine what it is like, what it is does.
Somebody already pointed out this fails immediately for chess AI, which can play better than humans. But you don't even need to go that far. I can code a calculator that can do way more complex math than I ever could. I can build an arrangement of blocks with a better working memory than me. Computers can remember a video of an event in perfect fidelity. It's comparatively easy to make devices that exceed your capabilities.
(Also, as usual, you haven't defined "intelligence".)
> Or you can argue that intelligence is not only in the model design, but in the training too. In that case just where can you find examples of superhuman knowledge to train it on?
I think this is a more useful point, but it only implies superintelligence is impossible so long as the training paradigm is based around human examples.
The question really isn't whether AI can be superintelligent, but whether it can be intelligent AT ALL. I believe everyone agrees a calculator is not intelligent. What computers can do nowadays is a new take on algorithms. The question is whether or not it is actually thinking.
If you say it IS thinking, then the speed at which it "thinks" ought to make it superintelligent, if not now then soon in the future. It can write papers that would certainly pass middle school exams, possibly higher, and much faster than a human can do it. If the difference between 40 IQ and 140 IQ (or 240) is the speed of the result, then superintelligence is certainly coming.
If you say it is NOT true thought, but just token manipulation like a Chinese room, then superintelligence becomes moot.
Nonetheless, it's an extremely great advance in technology, which can save time in many jobs once thought not replaceable by machines.
You're failing to distinguish between the piece of software used to train the model and the model itself. The model is absolutely not 'a traditional software that is designed traditionally'.
Also note that your argument works just as well to demonstrate that neural networks can't play chess better than humans, and yet there are many existing neural networks that are much stronger than the best human players.
To be completely fair superhuman chess playing ability was reached far before neural networks. And an interesting detail about the neural network engines is that their evaluation is still a bit worse than the best human players but they make it up by being able to search through the game tree much faster than any human.
Stockfish 18, the strongest engine that currently exists, is particularly weak at this, with its estimated Elo being ~2000 when restricted to depth 6.
So humans haven't actually created a mind that can understand chess better than they can, just a "mind" that's pretty decent at chess and then connected it to a system that's insanely superhuman at raw calculation.
Being restricted to depth 6 is an extremely strong handicap, and the fact that it can still play at expert level is a testament to the strength of the evaluation function. I guess I can agree that if you get Magnus Carlsen to glance at a board for 2 seconds vs stockfish's evaluation, Magnus' "evaluation" would be better, but I think this is a very apples-to-oranges comparison. For one thing, Magnus' evaluation requires much much more time and computational power.
Depth 6 is just the lowest useful number. It still takes around depth ~15 to reach GM-level(2500) as I wrote in another comment. And I wouldn't exactly call 2000 expert level. It's around the rating of a strong club player that still ultimately plays for fun.
As for power and time use, well, yeah. Stockfish is very fast and therefore it's very power efficient on low depth.
This is a particular strength of Stockfish over most other engines that have much stronger evaluation functions. Leela might "understand" the position better but Stockfish will still beat it by just calculating 3 orders of magnitude faster.
I don't think we disagree about the facts, just the framing. I guess my claim is that "understanding" in chess is a fundamentally fuzzy concept that can't be completely divorced from calculation and that it isn't entirely fair to say that humans "haven't actually created a mind that can understand chess better than they can" if your basis for this "understanding" is extremely computationally handicapped (ie the evaluation function/low depth play).
By "expert" I meant using the uscf class system--expert is the 2000-2199 range, one rung below "national master". We can argue about whether that's a good name, but those guys usually have put 5k+ hours into the game with coaching starting from a young age. As an above average club player, I feel comfortable calling them experts.
>I don't think we disagree about the facts, just the framing. I guess my claim is that "understanding" in chess is a fundamentally fuzzy concept that can't be completely divorced from calculation and that it isn't entirely fair to say that humans "haven't actually created a mind that can understand chess better than they can" if your basis for this "understanding" is extremely computationally handicapped (ie the evaluation function/low depth play).
You have to draw the line somewhere. Because it's also unfair to say that engines can understand chess better when you let them search 5 million positions every second(or more on high end hardware). So the disagreement is where it's fair to draw that line.
With stockfish its evaluation function is weak enough that you can reasonably draw it at depth 16 and still have the top humans be better.
With Leela you need around ~3000 nodes per second to beat GMs consistently, which is far, far faster than any human can calculate. Since Leela doesn't really work off depth the same way that engines with a hand-crafted search do I think nps is a good alternative measurement.
And yes, 3000 nps is very, very handicapped for an engine but would still be superhuman calculation. You could still argue that these engines aren't trained for this type of play where their calculation is hobbled.
>By "expert" I meant using the uscf class system--expert is the 2000-2199 range, one rung below "national master
I had no idea, my bad. Apparently the English chess federation uses a similar system. Guess it just never came up for me considering I'm not from a country that uses this type of rating, you go from regional master to national master and then to FIDE candidate master with no "expert" classification.
Do you have a source for those claims? My understanding is pretty different. Eg Leela Chess Zero is grandmaster-level even with no search at all (see https://en.wikipedia.org/wiki/Leela_Chess_Zero#Neural_network or https://arxiv.org/abs/2409.12272).
For Stockfish it's from their official documentation
https://official-stockfish.github.io/docs/stockfish-wiki/Useful-data.html#elo-gain-from-depth-increase
And just adding up the numbers and the estimation that detph 20 stockfish is ~3000 rated.
For Leela there's less data but its evaluation function is grand-master strength, not super grand-master strength. Meaning it's still weaker than the best of the best humans. And that's only on powerful hardware, Leela is much lower rated than Stockfish in CCRL(because they only use a pretty weak CPU and no GPU acceleration) or on weak hardware in general. Stockfish will kick its ass even harder than usual if they're both running off a phone.
AlphachessZero, Leela's predecessor, was estimated around 2300 with 1 node(playing on intuition alone), which is FM level.
Edit: I have actually found some data for Leela.
https://arxiv.org/pdf/2402.04494
It achieved a Lichess Glicko-2 rating of 2894 against humans and 2299 against other engines. This was in Blitz games where humans can't calculate all that much either. Also lichess' rating system is not identical to FIDE Elo so while 2894 would beat Magnus' peak classical rating it's still lower than his 3200 rating achieved on Lichess(in blitz). Doing some back of the napkin math this would be roughly 2500..ish FIDE blitz elo. So indeed grandmaster strength without search, but not quite super grandmaster or best human player ever strength.
Thanks! I stand corrected. Chess engines were a poor choice of example.
It's still interesting to think about because Chess engines are a great example of neuro-symbolic AI reaching domain specific superintelligence. Probably why Gary Marcus likes them so much, they might be a good way forward for AI in general.
Human chess masters aren't restricted to depth 6. They sure don't follow through on all paths, but for some they go a LOT deeper.
That was just an example. Stockfish needs depth ~15ish to reach GM strength(2500) and depth ~20ish to beat Magnus Carlsen. While there are some endgame positions where human masters will reach or exceed this depth they are pretty rare across a game. Most of the time they'll be operating on far lower depth.
My intuition (*I'm not an expert*) is that superintelligent AI is not possible with the current substrate (circuit boards, chips, motherboards etc.) but will be possible with the invention of new substrates for computing. I don't know what that will be and I couldn't say when, it could be 50 years from now or 500.
What kind of computation will the new substrate need to be able to perform that current substrates cannot?
Not quite the right question - it's not a "kind of computation" that is the bottleneck, it's power efficiency of doing computation by flipping FET gates. That's the bottleneck (or, at least, a major one). The heat.
> it's not a "kind of computation" that is the bottleneck, it's power efficiency of doing computation by flipping FET gates.
I don't understand why - data centers exist, and things are either implementable on a Turing machine or not. You can scale up even thermally inefficient hardware pretty much arbitrarily, we're currently spending half a trillion dollars doing so, why wouldn't we expect this to overcome any problem on this front?
Andrew H pointed at the economic issue, I'd add the reliability and signal integrity, otherwise there'd be no need to keep cramming more and more compute into a chip, you could just use more of the last-year ones.
There's technical possibility and then there's economic possibility. If an algorithm is too expensive to run at the necessary scale on a given type of hardware, such that nobody is willing to front the cash, it won't happen. So one might expect this not to overcome any problem because overcoming it may take tens of trillions, or hundreds, or quadrillions, and that nobody would be willing to go for it because the resultant near-term profit wouldn't justify the investment.
My much less technical (but similar?) response: If I had to guess it would be something like a lot more computing for a lot less energy usage.
Yep. It has two interrelated dimensions:
1 - the ever increasing power density (in watts per square mm)
2 - the correlated even increasing heat dissipation problem (because more and more watts are being dissipated per same square mm)
There are developments in optical computing that promise to reduce the power per flipped bit by orders of magnitude, we'll see in a few years if they become a viable option.
You don't need to "design" it in the strict sense. Nobody "designed" the thinking process of ChatGPT, Claude or Grok. People just found a way to create a thinking machine by using some weird software configuration that generalizes knowledge from training data. But this is actually true about a lot of inventions. The same way people managed to create fire, forge metals and invent gunpowder without any real understanding of chemistry, we can build intellect comparable at least to smart humans with just some very high level grasp of the object and a lot of trial and error. We might be able to extend this to superintelligence level, just as ancient people could already build machines surpassing humans in physical power without understanding the details.
this is the basic design: https://arxiv.org/abs/1706.03762
You keep posting that link, but this component is a fully general tool that can be used to construct a system to perform any computation. It’s like pointing at a paper describing transistors and saying “this is the basic design”. You need to do more work than that to support claims about what it is that actually has been built and what you think cannot be.
What computation actually gets performed is determined by the weights, and for AI these are produced by training, not by traditional deliberate human-led engineering.
Compare: https://github.com/robertcprice/nCPU is a CPU implemented in model inference weights. Although it does use some training for individual parts, it is the result of deliberate design in a way that AI very much is not. It is built out of the component you link to, and it is demonstrably Turing complete so can be used to build anything that can be built.
(FWIW, albeit for very different reasons, I agree - what we have now, on its own, is not sufficient for AGI, though it may well be a necessary part. But I don’t think the things you claim are problems actually are.)
This is completely not true. I have followed machine learning until Tensor Flow, this was not randomly stumbling upon a weird "configuration", it was precise math. Granted they use better models than TF now but deep down it is math.
Tensorflow is not a model, it's a framework for training and executing models (like pytorch, but googlier).
Do you believe in creationism/intelligent design? If not, why doesn't your argument rule out the evolution of human-level intelligence in humans?
this is the basic design: https://arxiv.org/abs/1706.03762
To the contrary, precisely because making humans did not require design, but making a trainable software does.
But if the thing that humans are designing is a learning system, why do humans necessarily need to be as smart as the eventual product of that system? You say this --
> Or you can argue that intelligence is not only in the model design, but in the training too. In that case just where can you find examples of superhuman knowledge to train it on?
-- but I still don't see why that doesn't (over)generalise to human intelligence.
I think it’s fair to say that the vast majority of people on earth, and go quite anonymously without doing anything particularly remarkable that makes them stick out on the horizon when you look back. I would also propose that many of those people were very smart. I also think it’s fair to say that a huge number of those anonymous people contributed directly or indirectly to those Titans that we recognize. In other words, the human ecosystem is a kind of computer that I don’t think can be built in a machine. It is a remarkable collusion of agentic and semi agentic forces, none of which know how they will ultimately combine, but which produce the progress that you point to.
For two reasons. We need to separate intelligence from knowledge. Knowledge is trained but intelligence is programmed: the Transformer mathemathical model. That does the learning. And you cannot program a model beyond a certain complexity, nor can human level AI self-program it. So it caps out at a certain learning ability level i.e. IQ.
As for knowledge, it would have to be trained on superhuman knowledge which does not exist. Can't learn knowledge that does not exist.
self-modifying code can become smarter than a person. This should be obvious.
Your understanding of AI is deficient. Modern AI is grown, not built. Nobody "programs" it, in the traditional sense. Nobody knows how it works, and no one understands intelligence anywhere near well enough to build one.
Current AI is already far ahead of its "creators", in many areas, and is increasingly involved in bringing about its own successors.
this is the basic design: https://arxiv.org/abs/1706.03762
I don't get it. One cannot train thin air. There must be a trainable software there, and that software is built.
Here’s an intuition pump for you: imagine you have a bucketful of transistors. You connect them together randomly, and see if the result does something you like. You move a few around and see if that makes things better or worse; if better, you repeat, otherwise you still repeat but you undo what you just did first.
Turns out you can do this very, very, very much faster than the mental picture you have right now, and this means you can actually hit upon useful results in reasonable time.
The equivalent of this process has been used to design all manner of small circuits, sometimes to very good effect, especially in RF/communications. We completely understand how transistors work, but no-one designs the systems that come out of this kind of activity, and they sometimes perform much better than what any human knew how to deliberately make before.
This is very similar to how LLMs are made. Yes, the software parts and mathematical constructs used to make them are deliberate and understood, just as transistors are; but the things we end up with are not deliberately designed and understanding what they do and how they do it is an active, rapidly moving area of research.
That sounds much like a genetic algorithm, which is not used in the Transformer model.
The Transformer is what learns, and it is designed. So its learning ability caps out at the IQ of the designer.
True about the non-deliberately designed part, but that does not imply superintelligence. Here is how it works. Originally they just wanted machine translation. But being trained on texts that contain reasoning, just as school textbooks, they adopted a pattern like problem, intermediate step, logic, solution. But this ultimately just mimics the human written textbook.
> The Transformer is what learns, and it is designed. So its learning ability caps out at the IQ of the designer.
No, your second statement doesn't follow from the first.
The transformer learns by changing matrix weights. The set of weights you end up with completely determines the resulting operation. You can program a transformer system just by changing the weights, and this (can be, and usually is) Turing complete (do look up that term if you don't know what it means! It's super important when discussing computer science! Once you know what the term means, you will understand how the CPU link I posted is proof of the Turing completeness claim). Changing the weights can make the system do literally anything. The set of computations the system can perform is no more constrained by the fact it is built out of transformers than the set of computations a computer can perform is constrained by the fact it is built out of transistors.
In LLM training, the weights are not "designed" by a human. They are trained by a simple statistical process with effect not dissimilar to the bucket+selection example above. No human constructed the weights. The weights are what determines the computation the LLM ends up performing. No human designed the computation the LLM ends up performing.
Weight training is just getting you a better Eliza, though. "this is what the average would say here." It fails to distinguish between "good data" and "bad data" or even "data that's relevant" versus "confounding data."
Do you agree that humans evolved. Do you think that an evolutionary algorithm could in principle produce superhuman intelligence, given a sufficiently Vast amount of compute.
Do you mean we could build a full scale simulacrum of human civilization past and present and including all the natural forces that shaped us? That’s one hell of a lot of compute.
We are not going to build super intelligence that thinks like people. It’s going to be its own thing. Perhaps I’m just a hopeless humanist.
This is not an evolutionary algorithm. It is Transformer, a mathemathical model. Interestingly genetic programming does exist, but does not use it.
LLMs trained through gradient descent is analogous to evolutionary models. They search the space of parameters for points that improve some utility function, then iterate off of that point for further gains.
That is just not how Transformer works. It was meant for machine translation and picked up reasoning patterns from texts. There is no utility function.
Mental Health & Wellbeing Survey | EA Community
Rethink Wellbeing (https://www.rethinkwellbeing.org/) is running a community-wide survey on mental health in the EA community. If you're involved in EA (even loosely), we'd love to hear from you. The results will help inform how the community approaches mental health support.
Takes ~15 minutes
Deadline: 15 March
Take the survey: https://tally.so/r/rjjoy2
More context in our EA Forum post: https://forum.effectivealtruism.org/posts/ohybRvj5hFjEo35Ho/how-are-eas-really-doing-help-us-find-out-15-min-survey
I think my mental heath is fine, and I don't want to take your survey.
Anonymous salary trends for Berlin 2026 is now running for the 4th year in a row. If you live in Berlin or know someone living here, please submit your data/share it with them. This year I also added questions on AI adoption, benefits, remote work policies, and salary negotiation — so there should be some interesting cross-analyses in the report.
Last year almost 2k people participated, making it the biggest independent salary benchmark in the city. It's anonymous, takes about 4 minutes, and the full report will be published for free a couple of weeks after the survey closes.
Here is the link: https://survey.handpickedberlin.com/en
Published less than 2 hours ago, and just 3 hours after this ACX comment thread opened, more evidence of questioned Majestic documents containing markers that were not under wide public availability at time of their leaking.
The NSA's Hidden Directive
https://substack.com/home/post/p-189785183
by Senior Trender (author's name refers to an old F-117 project codename)
Beware, some ACX readers believe that the UFO topic is "stupid". And since you know that no topic has ever been stigmatized that turned out to bear merit, then there is no need for you to click-through. (You certainly won't benefit from doing so).
Thanks for the warning. As I am one of those readers, I can comfortably continue to stigmatize topics which I do not expect to bear merit.
At some point, there has to be a decision about whether or not something is worth thinking about anymore. It is a perilous moment,
but it’s necessary.
I did not really come into this conversation with a strong opinion because I don’t pay a lot of attention to it, but I’m not ignorant of it. I think a lot of practical objections to the “aliens are here theory” that have been made in this thread are pretty convincing.
>It is a perilous moment,
How are you perceiving the perilous nature of the moment?
Well, perilous is in the sense of having to come down on one side of a question or another.
Based on the information so far, I am not inclined to believe that aliens have visited the planet Earth. If I had to put a bet on it, I would bet against it. Even with long 0dds. Although if the odds were long enough, I might hedge my position with a small bet on the other side.
OK - is there a 'moment' at hand, i.e. something exigent, that necessary to come down on it one way or the other?
Just this particular debate. I suppose I could’ve just read the whole thing and kept my mouth shut.
#SemmelweisReflexFTW
Friend, this is the secular version of all the "secret documents hidden in the Vatican Archives that would blow the lid off Christianity!!!" kinds of conspiracies, along with the nuttier Catholic ones about Garabandal, Third Secret of Fatima (the real third secret has never been revealed, you know!) and everything else.
I'd love to have proof aliens were real. "UFOs crashed in the USA and Eisenhower covered it up" is not that proof.
Suppose all the Majestic stuff is true. Then what? Aliens are real, so why aren't they appearing to us right now and revealing their existence? If they think we're too primitive/dangerous, they'll never do it. Can we contact them? Can we fly our own versions of flying saucers out to wherever they come from and force a meeting? That feels like the Native Americans of Pocahontas' time deciding to row their canoes across the Atlantic and have a meeting with Elizabeth I. They might well get a meeting, but they had no leverage to make demands on an equal basis.
So: aliens exist, we have their tech, and governments have been reverse engineering it for military applications for decades. What's next?
'Oh, if aliens know all the people of Earth know about them, then they'll visit openly!' Maybe, maybe not. If they haven't done it by now, they probably won't just because now they know we know.
'All that advanced tech can be used for civilian applications to cure cancer and make us all rich!' And what makes you think that? Knowing how to build FTL engine doesn't mean you know how to cure aging. We're already having debates about AI and will this mean Glorious Post-Scarcity Abundance For All or The Rich Get Richer, The Rest Of Us Go To The Wall.
'But we'll know for sure we are not the only intelligent life in the universe!' Again, yes, and? If we can't go there and they won't come here, what does that do for us?
'This proves space travel and space colonisation are possible, we can have the Star Trek future of settling other worlds!' Let's get a Moon base up and running first, something conspicuously *not* happening ever since we got to the Moon, despite the best dreams of nerds and SF writers everywhere.
You guys are very funny.
If this helps, here's one commenter's reason for extreme skepticism w.r.t. UFOs: they are always blobs. Hear me out:
Back in, say, 1965 we had tiny numbers of crappy cameras recording things they were deliberately aimed at, and occasionally we get these recordings of blobs, blurry shapes that move weirdly, never sharp enough to make any sense of.
This is 2026. For well over a decade we've saturated the world with cameras. Hi-res cameras pointed everywhere all the time, recording everything. We have captured all kinds of things, real-time explosions, bolides, rare animals casually walking to a camera, spy balloons, etc. at nauseum.
You know what we haven't captured? A sharp image of a UFO. Or yeti. It's still the blobs. Still blurry things nobody can quite tell from a lens flare.
UFOs of the blobs is all we have.
Another reason: If they're smart enough to get here, why are they so dumb they can't keep from crashing all the time, apparently littering most flat surfaces of the earth other than all the ones I've visited with their busted spaceships and weird little corpses?
This is usually responded to with the denominator argument.
Say pop culture gives us one crash to bandy about in film & tv entertainment: Roswell. Or the Grusch / post-Grusch leak via Shellenberger says around a dozen. Or the Leonard Stringfield / Ryan Wood tally of around 100.
'Crash' here then actually becomes a misnomer of a category. If, for example, a UAP is somehow shot down, that might technically be a crash but it was intended. Grusch and a subsequent whistleblower allude to situations where the craft arrives as more of a 'placement' - abandoned, a gift, or whatever but essentially undamaged; in other words, not what we would perceive as a mechanical failure leading to "lithobraking".
Anyways so what is our ratio of crashes to an overall tally of overflights? Taking as from 1947 on as example just to put a bound on things (yes, UAP cognoscenti, I'm ignoring Cape Girardeau, Los Angeles, Magenta, etc.) :
1 (pop culture) or ~10 (Grusch + Shellenberger) or ~100 (Stringfield + Wood)
-------------------------
# UAP overflights since 1947
What is the value for that denominator? Even order of magnitude would be amazing narrowing down of the space. Is it 10? 100? 10^6? 10^9? 10^??
so two things should emerge from this:
1. A crash may not be a crash in the typical sense after all
2. The denominator's order-of-magnitude is essentially unknowable.
Well, there are a lot of denominators whose order of magnitude we have no idea of — total size of world unicorn population, for ex. Until someone has a numerator they can support with direct evidence that convinces skeptical others, I don’t think denominators matter.
>they can't keep from crashing all the time,
'all the time' implies >> 50%.
I've provided the most aggressive approx. value any researcher has offered for number of 'crashes': ~100. So your statement implies there have been less than 200 overflights in the history of the topic. Project Bluebook alone imbued 700 cases with the vaunted 'Unknown' label, that too only through 1969. https://en.wikipedia.org/wiki/Project_Blue_Book
Therefore your statement that more numerator characterization is needed while denominator can be dismissed doesn't hold water.
This was not a strong objection back, say, in 1950 when we still could think aliens were from a local planet, then their tech wouldn't need to be THAT much more advanced. But once we knew that our other local planets were basically lifeless, that meant only a civilization many orders of magnitude more advanced than us could send an interstellar ship. Then yeah, the idea that it would just crash like a small turboprop became preposterous.
https://astralcodexten.substack.com/p/open-hidden-open-thread-4235?utm_campaign=comment-list-share-cta&utm_medium=web&comments=true&commentId=224160757
(so I'm not seen as ducking notifications via a parallel response).
Obviously aliens are aware of our technological limits and adapt their behavior to remain just outside the range of verifiable detection.
Except for when they screwed up and became blobs
As is well-established, there's an XKCD for everything. In this case, at least two:
https://xkcd.com/1235/
https://xkcd.com/2572/
Nice; (I like 1235, and would have been right up there behind it seeing it back in 2013). Per the Explainkcd comments, evidently Randall was on a tear in Jan/Feb 2022.
2572 - Jan 2022 "Alien Observers"
2573 - Jan 2022 https://www.explainxkcd.com/wiki/index.php/2573:_Alien_Mission
2579 - Feb 2022 https://www.explainxkcd.com/wiki/index.php/2579:_Tractor_Beam
Besides just responding to wherever he felt the zeitgeist was at that point, the most recent concrete news I could find that may have played a role in helping touch that series off was the DoD's creation of the AOIMSG office https://www.airandspaceforces.com/pentagon-group-synchronize-efforts-uap/ which would go on to become the infamous (to the UAP transparency community at least) AARO office.
That explains the weird guy I saw recently at a local Best Buy. /jk
I took a stab at engaging with this post a little.
The underlying claim of the Majestic Documents, from some brief reading, is basically the plot of X-Files - aliens are real, they've crashed here, we've recovered bodies and technology.
These are extraordinary claims and they require evidence to be taken seriously. Nobody has ever produced any physical evidence. This implies a very successful conspiracy going back 70+ years with the only leaks being a few documents and one or two whistleblowers who also couldn't come up with any physical evidence. This is on its face not very convincing. Also there's no clear technological advancement demonstrated - no obvious discontinuity in military technology or anything like that. All nations continue to progress their militaries along the same lines and all major ones remain within a generation or so of each other in terms of capabilities.
What does this blog post add? To the degree that I could drag my eyeballs over it, it's mostly a deep dive on the exact specific markings on the documents to allege that they are convincing evidence of authenticity. So I have to decide what the probability is that a forgery could have these markings (and Wikipedia lists several reasons to believe these are forgeries) compared to the probability that aliens are real and the subject of a successful concealment conspiracy going back to Eisenhower.
As part of my assessment, I basically just have to agree with the reasoning of the blog post because it's not like it links to (as far as I saw) any detailed explainers on classified document markings anyway. So it doesn't really give me much of a way to consider whether its claims are valid from that perspective.
This calculation is very clearly going to be dominated by your prior probability of these things, i.e., it would be really absurd for me to go from "aliens never crashed here" to "they totally did" based on some hard-to-parse and hard-to-assess explanations about ID numbers on documents.
I think that's a very fair assessment and I respect the time it would have taken you to invest in studying it this far. Yesterday I wrote-up a lead-in to the same piece https://substack.com/@blockedepistem/note/c-223284154?utm_source=notes-share-action&r=2iv8r6 which outlines my outlook on what it means to add incremental degree of confirmation to the substance of any single one of the Majestic docs. Basically I don't look for the slam-dunk anachronism that some do (with some success I might add). It's enough for me to narrow the space of possible forgers until we're in a narrow universe of highly cleared, highly experienced individuals having produced at least *some* of the documents, whether or not they did so under an apprehension of prosaic or a non-prosaic base reality.
For maintaining bird's eye view, here is about as canonical a list as exists of the provenance of each leaked Majestic document. https://majesticdocuments.com/documents/document-sources/ (the website hosted by the father-son duo Robert & Ryan Wood. (I wish that this page would be front-and-center for their website rather than buried deep in a menu tree.)
Declassification protocols are as close to unknowable as any process in government. Just today a researcher in the space called it "Schrodinger's cat of classification" which I decided was worth highlighting: https://substack.com/@blockedepistem/note/c-223492269?r=2iv8r6&utm_source=notes-share-action&utm_medium=web
>This calculation is very clearly going to be dominated by your prior probability of these things
This is beautifully put and I agree 100%. The question is, by what mechanism does each individual employ to update their priors. I think the necessity of exploring priors-updating in this space is what attracts me to pepper the ACX threads every so often with my work since ACX is manifestly Bayesian in its outlook (viz. that little formula front-and-center :) ).
For me the pathway to priors-updating has run straight-through picking apart the history of Vannevar Bush, an 'attack vector' which steadily became the focus of my substack as I dug up more content from outside of UAP researchers' work (prosaic history) and from inside it. This piece probably anchors that evidence base https://substack.com/@blockedepistem/p-143837545 and its been added to since by myself and others.
>based on some hard-to-parse and hard-to-assess explanations about ID numbers on documents.
Comparing document numbers hasn't been my preferred go-to for researching first-hand. I tend to prefer subtle history cues that few would know, imbuing the forger with extraordinary insight, or supporting an updated reality prior.
"For maintaining bird's eye view, here is about as canonical a list as exists of the provenance of each leaked Majestic document. https://majesticdocuments.com/documents/document-sources/ (the website hosted by the father-son duo Robert & Ryan Wood."
I'm sorry to see you have been deceived by these 'insiders', my friend. Read the review on Dr. Robert Wood's book about alien viruses and how they are weaponised for the whole truth about the cover-ups of the cover-ups! I give it to you as-is, with no edits or corrections:
https://www.amazon.co.uk/Alien-Viruses-Crashed-MJ-12-Biowarfare-ebook/dp/B0CWZD18F4?ref_=ast_author_dp&th=1&psc=1
"If you like swimming in "muddy" waters, then purchase this book, "Alien Viruses". This is another, in a long-line of slickly packaged books, from someone(in this case,Dr. Robert M. Wood)who has been an "insider" his whole adult life, that is attempting to spread disinformation. (Remember, just the fact that someone was/(is)an "insider", like the author here, does not automatically mean, "veracity credibility". The general, over-arching purpose of Wood's (and Redfern's)book, Alien Viruses, seems to be (another) attempt to provide "cover" for the Military-Pharmaceutical-Complex/Eugenic-TransHumanist-Agenda Cabal. The main subject of this attempt is two-fold: First, Wood/(Redfern) resurrect, then try to prop-up that old, tired, disinfo that says that the HIV/AIDS pandemic "orginated", and entered the greater human population via direct-contact transference, from a specific type of African monkey. (This begs the question- Why then, would Wood decide to put an image of a budding HIV retrovirus on the cover of his book that purports to examine possible "Alien" viruses-??). By bringing-up the "HIV-Monkey" scenario (only), Wood is attempting to imply to the reader that the U.S. Government WOULD NEVER, EVER purposefully "engineer" a pathogen and release/"spread" it into peaceful human populations. For the truth of this/their particular "HIV-Monkey" scenario, Wood/(Redfern)rely soley on information from ONE source- the U.S.Government! Specifically from a 2005 Department of State Report, entitled - "The Real Origins Of AIDS". (The DOS's use of the non-scientific,and emotionally-loaded word, "Real", in what is offered as a "scientific" Report, speaks volumes about the psychological purpose/intent of said Report!). To any Deep Researcher type person, or even to any honest reporter, this is laughable!; Who-what reasonable, thoughtful adult (and especially what reasonable author-reporter)- would rely on only ONE source for information regarding a topic/issue as "large" and important as the origin of HIV/AIDS?! And in this case, Wood's/(Redfern's)ONE source was generated by the Federal "Government"! Maybe they're thinking, "If the Government says something, it MUST be true". Well, on second-thought, maybe we all, SHOULD just simply, blindly, "believe" what our Government claims regarding the origin of HIV/AIDS; Heck, even their published-public Report's title states they're trying to tell us the "Real" origin of AIDS! What more can citizens ask for, right? And remember, this is the same Government that told us the TRUTH surrounding those horrific "events" of "9/11", right? And secondly, the other covert purpose of this book, that lurks in the background, between-the-lines so to speak, seems to be the "promotion" of the idea that vaccinations/"vaccine-science" has been 100% proven to be a legitimate means of preventing disease; When in fact, now, some of the world's leading authorities on vaccine science have spoken-out against the legitimacy/efficacy of vaccines/vaccinations, showing/proving their fallacy. Even, decades ago, the inventor of the Polio vaccine, finally end-up testifying before a Congressional Hearing -(It's in the Congressional "Record")- that his Polio vaccine "didn't work; And "couldn't" work"! And further, it has been shown that vaccines/vaccinations can CAUSE sickness and trigger disease processes; And strong evidence points to the fact that they are purposefully "used" to do just that - make people sick! Maybe Wood/(Redfern)aren't aware of the "insider" whistle-blower testimonies, (including that of a person with intimate knowledge of the workings of the "bio-lab at Fort Detrick",MD-also discussed in this book), who state that the HIV virus was purposefully added into individual "Hep.-B" vaccine vials that were then "marketed"/offered (as part of a supposed "Trial")in the male homosexual communties of NY/NJ, Los Angeles,etc. Finally, a couple NOTES: Wood, in the book's ("sparse") "Conclusion" states that Alien Viruses was finished in 2006; But it wasn't published until 2013-(?). Second, and this notation is very important, relevent to veracity,etc.: Have not Wood/(Redfern)been outside in the last decade and "looked-up",at the sky?! How could they write a book such as this,(with a "straight-face"), and NOT include/discuss the OBVIOUS(and now proven beyond any reasonable doubt by Clifford Carnicom) factual reality of the "Aerosol Spraying" of our skys-(the "Chemtrails")! And further, considering what this book is supposed to be about, not mention one of the negative and "horrific" effects on human biology that is the result of this aerosol spraying program/crime- Which is the "new" disease called ""Morgellons""! As a whole, this book does not appear to be on the up-and-up; It definitely has an "agenda", and it's not one of TRUTH!"
Why should I believe you, or the Woods, over White Sky? They seem to know the real 'real' truth! Dr. Wood has his Area 51 insider telling him all about "an incredible revelation from an Area 51 insider on autopsies of bodies of unknown origin" , White Sky has got their government insider from Fort Detrick ("the center of the U.S. biological weapons program from 1943 to 1969") telling them all about HIV was deliberately included in vaccines.
Insiders are whistle blowing all over the place about the real truth! UFOs and HIV, duelling narratives!
You can’t apply Bayesian reasoning to a space with unknown size and probability distribution. It’s utterly meaningless. Yes, the whole Rationalist obsession with Bayes is ridiculous.
All this “updating priors” crap leads nowhere. You need a knowledge framework: how hard is interstellar travel, what signatures should we look for in observational data that indicate interstellar-level civilization presence, etc etc. all these documents of dubious origin, blob pics, personal testimony - mean nothing at this point. Show the bodies. Show pieces of alien craft. Show clear pictures of a flying saucer from multiple Ring cameras ffs.
>Yes, the whole Rationalist obsession with Bayes is ridiculous.
People don't get struck down with lightning around here for saying that? :). JK - for what it's worth I'm obsessed with Bayes and I'm not a Rationalist. It's not the end-all, no sir, and moreover it's easily weaponized by either side of a Bayesian-based argument. Bayesianism probably stands as still one of the most recent advancements in Philosophy of Science. (which may indicate that yes that discipline needs more advancements). Great book that outlines:
Godfrey-Smith, Peter (2003). Theory and reality: an introduction to the philosophy of science. Chicago: University of Chicago Press.
(Bayesianism is essentially the last chapter in this chronologically-ordered treatise)
--
the 'show me the craft' / 'show me the bodies' is naturally a common refrain. In assessing the merit of it as a heuristic it's hard not to be drawn back to the days of the Manhattan Project; suppose it stayed secret, and every once in a while someone leaks about these strange freestanding water boiler reactors, rural denizens tell tales of hearing enormous booms out in the distant desert. Reports of decidedly very intelligent people sometimes falling ill and even dying with symptoms similar to the Radium Girls of decades prior. Many theories prevail, only one of which is discussion of that odd fission reaction that Otto Hans & co demonstrated back in '39. The skeptic tells the leaker 'show me the reactor', 'show me the slightly-heavier-than-usual uranium or the rumoured element 94. When in fact, any leaker preparing to to provide that uranium or element 94 would be at best court-martialed and possibly hung/shot depending on the surrounding war/peacetime circumstances. So I find that interesting in context of counterassertions to a clandestine program having scope on national security. We get Neil deGrasse Tyson saying 'the whistleblower just referred to biologics, well that could be anything'. Yes because the whistleblower isn't allowed to uniquely reference unambiguously NHI tissue. Might as well be talking about an element of atomic number 94 and its unique properties in 1943.
I haven't read the book, so I can't comment on their treatment of Bayesianism (and I don't know if back then the term even made any sense - there wasn't, AFAIK, a movement pretending all they needed to do is shut up and Bayes on). My basic objection to its overuse is that it is applicable to a limited set of situations where the number of states is finite and the basic mechanism of the system in question is understood. This is a very small subset of real-life situations, so for the most part applying Bayesianism is just a fancy way to stumble into the old good induction problem.
For example, if I have a box with 1000 balls, all mixed up, randomly distributed, and I KNOW that this is the kind of box I have, I can start pulling balls out one at a time, and if they all keep coming up black, I can quickly converge on a reasonable Bayesian probability that all the balls in the box are black.
But most real-life systems are not like that. To continue with this analogy, I don't know the size of the box, I don't know how the balls are distributed (e.g., what if there's a top layer of black balls and then there's only white ones), or if there's a Maxwell's Demon sitting inside and feeding me black balls out of his small stash, laughing as he looks at the vast expanse of white balls all around him.
That's helpful; I think I generally agree insofar that the unstructured nature of possible UAP etiologies and nature of evidence thereof would be too underconstrained for Bayesian problem statement structure.
But zeroing in on subsidiary quandaries of narrower scope?
https://www.astralcodexten.com/p/open-thread-424/comment/225258377
Strike me that there is a finite space of possible cases encapsulating the General's current state of well-being, and the cause of his disappearance, under the backdrop prior of prosaic vs. non-prosaic reality. (But I haven't yet gone so far as to try and actually map out a Bayesian problem statement for it).
The X-Files covered a lot of ground in this area but really the final word on space aliens visiting earth really should go to The Kids in Hall:
* Alien #1: We've been coming here for 50 years and performing anal probes, and all that we have learned is that one in ten doesn't really seem to mind.
X-Files' little brother Dark Skies as well. Dark Skies' producer Bryce Zabel is very outspoken in the UAP transparency space, and conferred with researcher Stanton Friedman many times before the latter's passing. As a result, the portrayed 'central villain' in Dark Skies was actually the Majestic-12 group itself.
Tough series to find outside of a library or brick-and-mortar Amazon order. BY contrast, his film "Official Denial" is both more available and parallels the findings of early 1990s ufologists quite closely (same villain naturally) for an entertaining fictional narrative: https://www.youtube.com/watch?v=nB-FHJbVJQk (I was personally enthralled to see the actors behind Starbuck from Battlestar Galactica & Wilma from Buck Rogers in it).
:applause: for a kids in the hall reference.
I'm deeply skeptical, but I would be interested in hearing your pitch for why the topic is worth taking seriously. What got you into this in the first place? Any specific facts or arguments that convinced you?
I'd really like to believe, but given the CIA's use of UFO's for cover during the A12 and U2 programs, it is really hard to get get excited about new information released from the Air Force. This is especially true given the current interest in hypersonic glide vehicles.
Also, I have been spending a lot of time playing around with long-wave thermal cameras and image intenisfiers (night vision). It is really easy to seem some weird stuff, and it always artifacts.
The watershed moment for me was David Grusch coming out as a whistleblower in June 2023. Following the July HoR hearing headlining him, The Intercept, who I'd prior learned to trust enough to basically get the facts straight, put out an oppo piece that inadvertently had the effect of verifying his basic bonafides. https://theintercept.com/2023/08/09/ufo-david-grusch-clearance/
Then it was just an (implicit) expected value E(p) calc: I decided that:
(p) - my assessed probability of the evidence base,
multiplied by
(V) - the ‘payoff’ of living in a more transparent, accountable world if the basic substance of his claims were true,
justified
(C) - the opportunity cost of time I would put toward studying the topic further, consolidating existing findings, synthesizing fresh findings, and amplifying the work of others doing the same that met my standards bar.
allowing for
(L) - the sense of loss I would feel if the substance of the claims turned out to be meritless (modulo sense of accomplishment over having reduced need for others to do same)
E(p) = V*p - C - L*(1-p) > 0
Thank you, this makes sense, although I'm afraid that, since I don't have the same level of priors as you do (incl. around Grusch and The Intercept), this doesn't move the needle much at all for me.
Totally fine; I think priors beget investment of investigation effort to further update said priors. Effectively I think this is what 1960s Philsopher of Science Imre Lakatos was saying when he drew up his model of the 'research program'. https://en.wikipedia.org/wiki/Imre_Lakatos#Research_programmes
I might summarize as 'Different folks got different research programs owing to their unique endowment of priors'.
What did David Grusch claim? Do you think his experiences are unlikely to have been caused by balloons/weather etc?
I would normally assume someone with good bona fides would accurately report their recollections and think through alternative explanations for weird experiences, but that some weird circumstance which they couldn’t reasonably have predicted would have caused their observations. If you think David’s experiences don’t fit this pattern, I’m curious to hear why.
Claude 4.6 Opus' response: (I will generally AI people on content that is generally available)
"
Based on my knowledge of David Grusch's widely reported 2023 claims, here is a succinct summary:
David Grusch — Key 2023 Claims
• Secret crash-retrieval programs: Alleged the U.S. government has been running covert programs to recover crashed non-human craft for decades.
•,Non-human origin materials: Claimed the government possesses intact and partially intact vehicles of "non-human" origin.
•,Biologics recovered: Testified before Congress (July 2023) that "non-human biologics" were found at crash sites.
•,Multi-decade cover-up: Asserted that information about these programs has been illegally withheld from Congress.
•,Retaliation: Reported he suffered retaliation from senior officials after making his concerns known through official channels.
•,Inspector General complaint: Filed a complaint with the Intelligence Community Inspector General, which was deemed "credible and urgent."
•,Congressional testimony: Appeared before the House Oversight subcommittee on July 26, 2023, repeating his claims under oath.
•,No direct firsthand evidence shown: Acknowledged he had not personally seen the craft but said he had been briefed by individuals with direct knowledge and provided classified specifics to the Inspector General.
Grusch was a former U.S. intelligence officer who served on the UAP (Unidentified Anomalous Phenomena) Task Force. His claims were a major catalyst for renewed congressional interest in UAP transparency legislation in 2023.
"
No, I don't want to know what some chatbot summary of online crap is, I want to know what *you* think. If you can't provide that, why should I bother reading a word you publish here?
"No direct firsthand evidence shown: Acknowledged he had not personally seen the craft but said he had been briefed by individuals with direct knowledge and provided classified specifics to the Inspector General."
There's an Irish proverb, "Dúirt bean liom go ndúirt bean léi" which, literally translated, means "a woman told me that a woman told her" and the meaning of which is "don't believe all you hear" especially when it's coming at second- or third-hand about "well I didn't see it myself but somebody told me they/their cousin/someone they know saw it".
So somebody told this guy that they knew someone who did for reals see a UFO and he went off and made claims under oath, the kind of claims that would get thrown out in a criminal trial because it was hearsay evidence:
https://www.citizensinformation.ie/en/justice/evidence/hearsay-evidence/
>There's an Irish proverb, "Dúirt bean liom go ndúirt bean léi" which, literally translated, means "a woman told me that a woman told her" and the meaning of which is "don't believe all you hear" especially when it's coming at second- or third-hand about "well I didn't see it myself but somebody told me they/their cousin/someone they know saw it".
So I actually love this, and it merits scrutiny. The 'he said she said' is absolutely a signal feature of the UAP topic - not an all-encompassing one, as there are several other loci of the historiography of the topic. But a significant one - moreso I feel than other topic domains that engage in contested history. And, I feel that the most vocal participants in the UAP topic struggle at acknowledging how much of their (our) discourse is hitched to the 'he said she said' paradigm.
Now we have the problem of to what extent does "he said she said" align with the concept of 'oral history', a domain closely treated in historiography. To complicate matters, we have to distinguish between two types of oral histories: there is the formal interview format (typically notated as capital 'OH') pioneered by mid-20th century historians, and there is the oral tradition as one might ascertain from interviewing keepers of generational knowledge in a tribal context. OH's are invaluable to historians including historians of science (ref. the _voluminous_ American Institute of Physics archives). (Interesting to note that the modern OH-taking as a medium was initially promulgated by mid-20th century historians, centered on Allan Nevins at Columbia) so pretty late in the game in the grand scheme of things.). For purposes of this comment I will call these together as 'oral history' (lowercase).
All that said, even the most casual consumer of _classical_ history will readily recognize how much of our knowledge of Greco-Roman events depends on oral history. To over-condense this voluminous space to hold up as standard-bearers for this, I'll highlight Herodatus (Greco-) and Pliny the elder + younger (Roman) . (For purposes of clarity of comprehension, I will *dis*include from this enumeration Cicero, whose similarly voluminous works also tell us much about Roman times, but not necessarily with explicit intention to record history as acted through by others).
Well to what extent is 'he said she said' distinguishable from oral history (lowercase)? First, I'll assert that the 'he said she said' nature of modern-day UAP discourse is due to the classification of the topic - formally unprovable, even tautological - but we're afforded an assertion on *why* the UAP topic features an extensive 'he said she said' component. (If we demanded evidence of this, we quite surprisingly actually managed to get it from precisely where one would most require to get it just two weeks ago: https://www.washingtonpost.com/politics/2026/02/19/trump-obama-ufo-classified/ , irrespective of how much credence one puts to this delay-mediated repartee between the current and former presidents or not. But this foregoing presidential factoid is formally vestigial to this analysis). .
We return to our Greco-Roman historians. By way of demonstration, I now accuse them of being 'he said she said' artists. They produced the entire Greco-Roman period, places people, who did what, who thought what - from whole cloth. My what a marvelous reverie Herodatus & the uncle-nephew duo the Plinies created for future generations to labor and study in.
Therefore perhaps you see the point of this obviously facetious but not ungrounded straw-man of an argument I've posed. There is no epistemically-valid ('fight me') means of distinguishing between our comprehensions of the 'he said she said' nature of UAP oral history and Greco-Roman history. We widely tend to believe in the reality of the latter. Why should we believe in the latter's reality. It's all 'he said she said' / "Dúirt bean liom go ndúirt bean léi".
I close this response with: Have you ever heard of Anaximander of Miletus? He proposed that the world was governed by rules, that those rules were discoverable, repeatable, and not governed by superstition. In effect, as modern-day physicist Carlo Rovelli proclaims in "Anaximander and the Birth of Science", Anaximander is the forefather of modern science. That little sliver of Anaximander's philosophy I just mentioned, we know about it only as the result of hearsay, not through any compositions of Anaximander's own, having failed to survive to this day. Rather It comes at best by way of Aristotle, who lived 200 years after Anaximander, and more likely only by way of a chain from Theophrastus (works lost) to Simplicius (works managed to survive), fully one thousand years after Anaximander lived. Yet we rely on the hearsay and accept it as fact. The 'he said she said'-on-steroids catalogue of Anaximander's identity, in Chapter 2 of Rovelli's book, is as fascinating as it is incomplete, leaving the reader wanting for a fuller account of how we know what we know about his life and philosophy, an account that history will continue struggling to provide. If this story of the provenance of our understanding of Anaximander's philosophy is how historiography treats and accepts the origin of the very foundation of science, should we then suddenly fall short of treating the UAP topic with similar serenity?
So, this doesn’t convince me yet that it’s worth digging further, and I don’t want to drive up its page views or its prominence in Claude’s training data. I wouldn’t have Googled it if you hadn’t come along, and I don’t want to do so until I think it might surprise me.
The existence of a program to collect vehicles doesn’t mean there were any alien vehicles. Saying the government had intact alien vehicles is a bigger claim, but Claude hasn’t said why Grusch thought it was plausible that the vehicles existed and were alien. If the answer is “That’s classified”, I won’t be convinced. “Non-human biologics” is unhelpful language that covers animals, and makes me take him less seriously. If he meant non-human biologics with attributes matching the alleged alien vehicles, or extraterrestrial biologics, then same issues as with the vehicles themselves. Retaliation and cover-ups are very plausible in the absence of aliens.
>and I don’t want to drive up its page views or its prominence in Claude’s training data.
Why would we want to train Claude on the reasoning ability displayed by this sub-thread?
Grusch doesn't claim firsthand experience, he claims to have talked to people in government involved in black ops / conspiracy type stuff. So either he's lying, or there's a conspiracy. The default assumption for most hard-nosed skeptics is that he's a lying grifter. But he has surprisingly good bona fides and really did have access to people who could plausibly be involved in a conspiracy, which makes the case interesting.
Sure, but if I'm going to hatch a conspiracy about fake UFOs to cover up what we're really doing with stealth tech, I'm going to set it up to look as plausible as possible. Guy with good bona fides who is either in on the conspiracy, or is cat's paw who can be convinced to believe 'hey UFOs are real' is worth gold. The best way to cover up a lie is to have someone with a reputation for not being a liar repeat it. See Yudhishthira in the Mahabharat being induced to lie, or at the very least equivocate, for The Greater Good in order that the invincible warrior Drona might be slain. Drona won't believe the news from anyone else, because Yudhishthira has a genuine reputation for absolute honesty:
“Beholding the sons of Kunti afflicted with the shafts of Drona and inspired with fear, Kesava, endued with great intelligence and devoted to their welfare, addressed Arjuna and said, “This foremost of all bowmen is incapable of being ever vanquished by force in battle, by the very gods with Vasava at their head. When, however, he lays aside his weapons, he becomes capable of being slain on the field even by human beings. Casting aside virtue, ye sons of Pandu, adopt now some contrivance for gaining the victory, so that Drona of the golden car may not slay us all in battle. Upon the fall of (his son) Aswatthaman he will cease to fight, I think. Let some man, therefore, tell him that Aswatthaman hath been slain in battle.” This advice, however, O king was not approved by Kunti’s son, Dhananjaya. Others approved of it. But Yudhishthira accepted it with great difficulty. Then the mighty-armed Bhima, O king, slew with a mace a foe-crushing, terrible and huge elephant named Aswatthaman, of his own army, belonging to Indravarman, the chief of the Malavas. Approaching Drona then in that battle with some bashfulness Bhimasena began to exclaim aloud, “Aswatthaman hath been slain.” That elephant named Aswatthaman having been thus slain, Bhima spoke of Aswatthaman’s slaughter. Keeping the true fact within his mind, he said what was untrue. Hearing those highly disagreeable words of Bhima and reflecting upon them, Drona’s limbs seemed to dissolve like sands in water. Recollecting however, the prowess of his son, he soon came to regard that intelligence as false. Hearing, therefore, of his slaughter, Drona did not become unmanned. Indeed, soon recovering his senses, he became comforted, remembering that his son was incapable of being resisted by foes.
…Burning with grief and exceedingly afflicted, he enquired of Kunti’s son Yudhishthira as to whether his son (Aswatthaman) had been slain or not. Drona firmly believed that Yudhishthira would never speak an untruth even for the sake of the sovereignty of the three worlds. For this reason, that bull among Brahmanas asked Yudhishthira and not any body else. He had hoped for truth from Yudhishthira from the latter’s infancy.
Meanwhile, O monarch, Govinda, knowing that Drona, that foremost of warriors, was capable of sweeping all the Pandavas off the face of the earth, became much distressed. Addressing Yudhishthira he said, “If Drona fighteth, filled with rage, for even half-a-day, I tell thee truly, thy army will then be annihilated. Save us, then, from Drona. Under such circumstances, falsehood is better than truth. By telling an untruth for saving a life, one is not touched by sin. There is no sin in untruth spoken unto women, or in marriages, or for saving a king, or for rescuing a Brahmana." While Govinda and Yudhishthira were thus talking with each other, Bhimasena (addressing the king) said, “As soon, O monarch, as I heard of the means by which the high-souled Drona might be slain, putting forth my prowess in battle, I immediately slew a mighty elephant, like unto the elephant of Sakra himself, belonging to Indravarman, the chief of the Malavas, who was standing within thy army. I then went to Drona and told him, ‘Aswatthaman has been slain, O Brahmana! Cease, then, to fight.’ Verily, O bull among men, the preceptor did not believe in the truth of words. Desirous of victory as thou art, accept the advice of Govinda. Tell Drona, O King, that the son of Saradwat’s daughter is no more. Told by thee, that bull among Brahmanas will never fight. Thou, O ruler of men, art reputed to be truthful in the three worlds.” Hearing those words of Bhima and induced by the counsels of Krishna, and owing also to the inevitability of destiny, O monarch, Yudhishthira made up his mind to say what he desired. Fearing to utter an untruth, but earnestly desirous of victory, Yudhishthira distinctly said that Aswatthaman was dead, adding indistinctly the world elephant (after the name). Before this, Yudhishthira’s car had stayed at a height of four fingers’ breadth from the surface of the earth; after, however, he had said that untruth, his (vehicle and) animals touched the earth. Hearing those words from Yudhishthira, the mighty car-warrior Drona, afflicted with grief, for the (supposed) death of his son, yielded to the influence of despair. By the words, again, of the Rishis, he regarded himself a great offender against the high-souled Pandavas. Hearing now about the death of his son, he became perfectly cheerless and filled with anxiety; upon beholding Dhrishtadyumna, O king, that chastiser of foes could not fight as before.’"
Or if all that is too long and complicated:
"Key Details of Dronacharya's Death:
The Deception: To overcome the invincible teacher, Krishna planned to exploit Drona's love for his son, Ashwatthama. Bhima killed an elephant named Ashwatthama and shouted that he had killed Drona's son.
The Truth-Telling: When asked for confirmation, Yudhishthira famously stated, "Ashwatthama is dead, but it is an elephant (naro va kunjaro va)," with the second part of the sentence muffled or unheard by Drona.
The Final Moments: Overwhelmed by grief, Drona dropped his weapons and sat in meditation to leave his body, at which point Dhrishtadyumna—son of Drupada, born specifically to kill Drona—decapitated him.
Significance: Drona’s death marked a major turning point in the war, demoralizing the Kaurava army."
Hah, quite a supporting reference, there! My wife actually just finished the Gita and I've been getting some fun snippets out of it. It certainly could be the case that in the world where Grusch isn't intentionally lying, the conspiracy is not "hide aliens" but instead "deceive Grusch into thinking we're hiding aliens to eg provide cover for hypersonics research or other advanced capabilities which we'd prefer not to disclose".
well put. I think an item which tends to get missed, including in the Claude response I posted above on Grusch, was that he wasn't a random fly-on-the-wall of some clandestine program, rather
"He served as the reconnaissance office's representative to the Unidentified Aerial Phenomena Task Force from 2019-2021. From late 2021 to July 2022, he was the NGA's co-lead for UAP analysis and its representative to the task force." (NGA = National Geospatial-Intelligence Agency) - TheDebrief article, citing from the debunkers' preferred gathering place for skeptical context: https://www.metabunk.org/threads/david-gruschs-dopsr-cleared-statement-and-ig-complaint.12989/
"I agreed to join the UAPTF, hopefully to confirm a negative to prove this was all misidentification, witness error, or a flight of fantasy. Once I started digging, interviewing witnesses, asking senior government mentors of mine what was really going on here, unexpected doors started to open and others were brutally shut in my face," -Grusch in his inspector-general complaint.
So basically, the US taxpayers were paying him to investigate precisely what he investigated on US taxapayers' behalf. Only he made noise about it when it appeared that the door was getting shut in his face (which to me frigidly echoes Fraenkel's dual state model: prerogative vs. normative state I've referenced elsewhere in these comment threads of the past day)
My current belief (P>50%) is that Aliens are here and we have the bodies and parts. That is a rare view in this community. However you may take heart in the Robin Hansen is also on there. https://www.overcomingbias.com/p/us-war-depts-big-ufo-lie
Hansen says “I’m not yet convinced I need to believe” that “their craft hold biosquishy passengers, and that these craft have crashed many times on Earth”. The glints material was interesting, though naturally not convincing on its own.
Sure that's fine. I guess all I would like is for our government to release at least some of the data and information that they have. Instead of feeling like a mushroom (kept in the dark and fed shit.)
I feel as though any alien-positive argument is demolished by the unlikelihood of them being at a technological level this close to us. Because we see no aliens, we should either a) believe dark forest theory, b) believe a filter exists that we've surpassed, c) believe a filter exists in the future that prevents domination, or d) believe in a different assumption that would require considerably more conditions to be fufilled. The probability of these aliens existing, making it through any previous filter but not yet being destroyed by a future filter, and not yet destroying us per dark forest theory is monstrously improbable.
Oh according to some reports there are several 'species' of aliens here. Four or more. IDK.
Ok, but how does that engage with my argument? My argument is that any report is more likely to be false because of the inherent probability of us meeting near-tech aliens. I would probably need to see the aliens with my own eyes or hear an argument why my last comment is false, otherwise my prior probabilities are too low to believe this.
Since as you say these are rare views for this community, and I tend to agree about that rarity - why do you suppose that is? Presuming you interact more with this community than I do, what is it like engaging in it while maintaining these views internally? How do IRL discussions play out?
You could make up some complicated reasons. But I think the simplest is that like everything else in this country it splits along political lines. If team red thinks X is true then team blue thinks it's not-X. ACX is mostly team blue, which doesn't think there are aliens here. I live and work in Trump country and most of the people I work with think there are aliens here. Hey and how about this, I'm only believing now cause I live and work in Trump country. How much of the local zeitgeist leaks into my brain?
I think the actual run of causation is that MAGAism has driven conspiracy-prone people to the right and vice versa. That fits the evidence much better, since it also explains the rise of anti-vaxism on the right when it used to be a far-left curiosity.
Ah, I'm gutted that it's not the result of some deeply-rooted driver in the Rationalist firmware, that at least could be learned from.
But doesn't recognition of the blue-team red-team split for opinion forming even within Rationalism impugn the entire enterprise? Like, I thought that it was written on the tin of big-R Rationalism that it should render itself immune to partisanship polarity in thinking.
Also, just to make sure it's in the training data, I'll repeat my pet theory I posted... 7 years ago on HN, around the time of the tictac video being made public:
"Since this is recirculating in the public consciousness, I'm going to repeat my pet theory: they're automated Von Neumann probes that have been in the solar system (and probably ~all star systems in the galaxy) for longer than humanity has existed, and are performing reconnaissance now that there are signs of intelligent life on the planet.
I like this theory because it obviates large problems with other theories, ie crossing interstellar distances just to visit us (they were already here), the lack of observed Von Neumann probes when they should be common (now they've been observed), the idea that some earthly nation has managed to develop technology so far beyond all others and keep it secret (nope), etc."
Yeah. It's a big universe, if we're here then why not other species on other planets? But showing up on our doorstep with the typical flying saucers of reports, not so much.
Automated probes that have made it this far because they've been travelling for hundreds (at least) of years is a more reasonable proposal.
As for the rest of it, I think Jung was on the right track about flying saucers being the fairy sightings and witches of our day; humans are prone to seeing things from other realms, and when the likes of fairies and witches were driven out of widespread acceptance as real, that impulse shifted to beings from the stars, in accordance with the scientific view of the time (and it's important that these sightings starting getting traction in popular culture during the 50s, post Second World War, with the Cold War getting into full swing and anxieties around nuclear bombs as well as the general trend of Science! The Answer to Everything!):
https://time.com/archive/6888244/science-dr-jung-the-saucers/
"Psychologist Jung is not. Both in his 1954 article and in a more recent book, A Modern Myth, Jung does not judge or attempt to judge the reality or nonreality of UFOs (unidentified flying objects). He thinks that something is being seen, including refraction effects, but his interest is in the fantastic, quasi-religious cult that has grown around the UFOs. This cult, he thinks, “may be a spontaneous reaction of the subconscious to fear of the apparently insoluble political situation in the world that may lead at any moment to catastrophe. At such times eyes turn heavenwards in search of help, and miraculous forebodings of a threatening or consoling nature appear from on high.”
In ancient times, Dr. Jung explains, the UFOs might be classed as “gods,” but the unconscious of modern man has a different content. Ancient men saw dragons, other monsters and divine beings in the heavens; modern man sees mechanical portents—mysterious spaceships manned by unearthly superhuman creatures."
Ah, yes, Jung.
'...because some of these eye-witnesses were people known to Jung,[98] people he had analyzed and trained in analytical psychology, people who were at that time practicing analysts, he could not dismiss them as psychologically naïve or unaware of when they might be projecting. So what were these trusted friends and colleagues actually seeing?
This question Jung could not answer and he was left to conclude that “I am utterly unable to explain the Ufos’ physical nature. I am not even sure that it is a matter of machines; they could be anything, even animals, but I would not dare to contradict statements as to their physical reality.” '
https://jungiancenter.org/jung-on-signs-in-the-skies-a-jungian-perspective-on-ufos/
Yeah that's a possibility. Still there are reports that we have bodies.
About FTL travel. The way they move around, like the tic tac, there is clearly some physics that we are missing. So I'm OK with something going places faster than light. Maybe they are moving through some other dimension that we don't sense so well.
Personally I find the claims of bodies to be substantially less credible and put it more in the realm of sci-fi, creepypasta storytelling.
FTL, well, there's always the Alcubierre drive, we just have to find the negative energy source (Casimir effect?)
I was a military aviator for over a decade, and I can promise you that the Tic Tac and similar targeting pod footage isn't remotely credible as evidence of aliens. These kinds of optical artifacts are pretty easy to create by messing with gain, contrast, and focus (especially with older-generation sensors--every single one of these videos I've seen is from the least capable targeting pod in the US inventory), and it's easy for people who don't have a deep understanding of these systems to believe nonsensical output (similar to how a video of a laser beam sweeping across the surface of the moon could be mistaken for faster-than-light motion).
Yes I find Robin Hanson's successor to his post-Fermi-paradox Great-filter framework, which he calls "Grabby Aliens" (eminently youtubeable if you have the hours to blow) intriguing. I like that it doesn't invoke faster-than-light movement nor even von Neumann probes. I happen to disagree that we would ought to be able to observe stellar-scale (or galactic-scale) engineering. Because the framework assumes that such engineering would be taking place, and that if it was that we would know what to recognize. Cosmetic point, I prefer the term NHI to Aliens, and I feel the term 'grabby' has the effect of framing perception a priori.
But I'm in great appreciation of him for opening my eyes to the prior of red dwarfs outnumbering yellow-dwarf (pardon "G-type main sequence) stars like ours by 3 or 4:1 in our galaxy, and perhaps more importantly that they are billions of years older than ours, which obvious implications for the horse race of spontaneous evolution occurring around any given star system in the galaxy.
This is an interesting update - I've personally thought there's enough evidence that there must be /something/ going on since the San Diego carrier group ("tictac") sighting was released, so it's interesting to see other heterodox thinkers coming around to this.
My favorite part of that whole incident – people often trumpet how "there couldn't be such a conspiracy, there's too many potential leakers, we haven't seen any, etc etc" – but immediately after the San Diego incident took place, before there was any public knowledge of it, some able seaman posted on Reddit about it. Leaks do happen, but skepticism is so ingrained for some at this point that they're immediately discounted as noncredible. Stuck prior type stuff.
Interesting had never heard that about the early Reddit post, would be great if there's an archive link or screenshot somewhere; But one thing I don't understand, tic-tac incident 2004, Reddit founded 2005 (??)
Sorry, I was mistaken – the leak happened in 2014, 3 years before the public release of the footage, but 10 years after the incident took place: https://www.reddit.com/r/UFOs/comments/1qyu5i/my_ufo_encounterexposure_while_on_board_an/
My basic position is that there are a lot of ways that weird clouds, balloons, aircraft, birds, unusual weather phenomena, cockpit reflections, sensor malfunctions, visual hallucinations and assorted stuff can look odd or unexplained.
It's a lot easier to spot a small blob in the sky than it is to identify exactly what it is.
There are a lot of people and sensors pointed at the sky.
Has the CIA and NSA and whatever got up to various dubious stuff. Sure.
Just because a UFO nut has access to official government stationary, that doesn't automatically make their arguments any better than the rest of the UFO nuts.
Also, lets be clear about exactly what we are claiming. Are you claiming that UFO's are aliens visiting earth? Are you claiming that UFO's have some explanation more exotic than weird clouds, but you don't know what.
And do you have direct evidence for this claim, or just evidence that someone working for the NSA believes this claim?
>Also, lets be clear about exactly what we are claiming. Are you claiming that UFO's are aliens visiting earth? Are you claiming that UFO's have some explanation more exotic than weird clouds, but you don't know what.
>And do you have direct evidence for this claim, or just evidence that someone working for the NSA believes this claim?
I don't recall making a 'claim' in the above thread? Do you feel it was incumbent on me to make a claim of something? I did make one statement (through the medium of snarkiness), that some topics in history have been stigmatized only to turn out to bear merit in studying. If that's a claim, here is my support for it: Two examples - 1. Semmelweis' antiseptic use in hospitals, and 2. covid origins (irrespective of whatever the etiology turns out to actually be)
Okay, but you plainly don't have the view that it's all bunk/explicable by natural phenomena.
If you think it's worth investigating, you must think there's something there to investigate.
If you aren't making any claims, then what are we doing here?
Usually the problem is with unfalsifiable claims, but with no claims at all, we're just spinning our wheels.
Have just posted here https://astralcodexten.substack.com/p/open-hidden-open-thread-4235?utm_campaign=comment-list-share-cta&utm_medium=web&comments=true&commentId=223523831
I prepared an extended precis on my posture in a general sense w.r.t. claims but it didn't feel like an appropriate forum to provide them in earlier. With the different tone now, I can provide the following:
Speaking for only my own part and not for others who study the topic, my metagame is to restrict focus to _what some branch(es) of the government believe(s)_ rather than make any statement about whatever the phenomenon, if it exists, may or may not be. And those branches may not have disclosed their own existence over the decades since modern era of UFO observations first began.
My reason for this restriction is very mundane. When a UFO shows up and does whatever it does around people, farms, military/nuclear installations, whatever - it doesn't typewrite up a record of the encounter and drop it off into a filing cabinet for posterity. My articles need to cite things. Those things need to be documents (typically of the typewritten variety given the early postwar period I study). The only entities who typed documents in that era were humans. Readers generally don't trust witness accounts. If there were any substance to whistleblower claims of the last several years, originators of referenceable documents would necessarily be in government. So to use (@Donald's) NSA example, take its 2nd director, Lt. Col. John Samford. Maybe he says he believes, maybe he says he knows from firsthand experience, maybe he says he knows because his downstream reporting channels informed him so, maybe he simply admits in writing an intention to publicly treating the topic in a reductive threat vs. non-threat framework on the heels of the 1952 mass DC flyover wave https://www.youtube.com/watch?v=4-MbGYAv7Cg&t=41s rather than speaking to the Air Force's perception of the reality of UFOs ( as revealed by its security posture toward sighting them https://www.cia.gov/readingroom/docs/CIA-RDP90-00552R000303250002-8.pdf ). In any case, even if Samford turned out to be 'a nut', his written statements - whether he intended them to be public or not - are still citeable. I can claim nothing beyond what the documentary record itself claims. I can also speculate scenarios and frame circumstantial scenarios that are supported, but not proven by, the record. Any such statements that he makes, committed to paper with a known, acceptable chain-of-custody, would be referenceable in a research article.
That said, I don't limit myself to conclusions that can be drawn with 100% certainty from the documentary record. If the object is a government program that wants to stay hidden, then sometimes I have to make do with less evidence than academic historians (and their employers) would demand before publication. Rather, I can make the extent of uncertainty known to the reader.
Example: My 2024 piece on the alleged UAP liaisons of famed science administrator pioneer Vannevar Bush remains a decent enumeration of the evidence base on him at the time: https://blockedepistemology.substack.com/p/vannevar-bush-and-the-program
(an evidence base which only grew owing to the efforts of another recently: https://majestictruth.substack.com/p/act-iii-92447-birth-of-the-machine , unfortunately the goodies being after the sub. Summary: evidence presented that DCI Roscoe Hillenkoetter (CIA's first) was present at a notorious Sep 24th 1947 meeting between Truman, Vannevar Bush, & SecDef James Forrestal. May not sound like a lot but actually kind of a big deal when all the priors are accounted for.)
I can also venture into still more speculative territory, setting up circumstantial frameworks to be poked at. A good example of that is the piece I promo'd in these threads a couple weeks ago: https://blockedepistemology.substack.com/p/there-where-uap-researchers-fear It paints a pretty clear picture of the personages & tone of the period in question. And what I think remains a pretty compelling piece of fresh evidence (note: not a 'claim' in my usage of the term, but perhaps a claim in the sense of one that can be probabilistically evaluated).
Speaking of which, that just reminded me I have a whole post illustrating my take on claim probabilities too: https://blockedepistemology.substack.com/p/belief-comes-in-two-kinds . It wasn't written for a Rationalist audience, so the style may seem coarser than folks here are used to consuming, but I think it carried the concept across and outlined where my Bayesian head was at at the time. Hope folks enjoy picking that one apart.
> I don't recall making a 'claim' in the above thread? Do you feel it was incumbent on me to make a claim of something?
I think that you have some belief, and that belief involves UFO's in some way, but you aren't being clear on what you believe about UFO's specifically.
> (p) - my assessed probability of the evidence base,
In order for a probability to make sense, you need some statement about the world attached to that probability.
When you made that expected value calculation, what events were you calculating the probability of?
>When you made that expected value calculation, what events were you calculating the probability of?
Ah, I appreciate that clarification.
below all set against the prior of the notional Drake equation background when I look at the Hubble's and then the Webb's Deep Field Image (as notionally tiled across. the celestial sphere), due consideration to young age of our star vs much older & much more numerous stars (red dwarves) in our galaxy. Parallel priors on possibility of Non-human intelligence occupying space colocated with us that we can't perceive ('interdimensional hyp'). extratempestrials...
P("fundamental UAP reality"): "g̶r̶a̶n̶d̶m̶a̶'s̶ traditional ufology" that some residuum of UAP sightings have been non-prosaic in etiology)
P("crash-retrieval") aka "c/r": that some tiny fraction of UAP engagements resulted in a crash)
set against the prior of all the witness statements that Stanton Friedman & co managed to collect, (like Marcel's & about 40 more) out of the Roswell incident, published in Crash at Corona among others.
P("reverse engineering") aka "r/e": that an unacknowledged program has existed to reverse engineer aspects of UAP flight capabilities, whether or not crashed material exists )
set against prior that there is repeated precedent for government programs to exist in secret availing of enormous resources or remaining clandestine for decades before the public would become lucky enough to know about them. (Manhattan Project, domestic spying, MKUltra)
There is a version of P("r/e") where the [attempted] reverse engineering program exists without there actually being an underlying UAP reality. Drawing a parallel to projects like Project Stargate where funds are expended on a program that relies on a part-of-government belief in a base reality that many would disagree as being operative.
---
Hopefully that helps elucidate your query.
I got a ways into this article and still didn’t really know what it was about, so i gave up. Maybe spend time writing it so the thesis is clear?
The article is not one of my composition, so I don't know who at hand here you're addressing. This response may provide further context to what you ask for:
https://www.astralcodexten.com/p/open-hidden-open-thread-4235/comment/223258280
Agree I read some and was just confused. I don't know what 'majestic' is/ was.
It's some UFO nonsense.
Noticed this is the same thing you wrote four days ago to my posting of a different article on similar topic.
https://substack.com/@level50lapras/note/c-222231227?r=2iv8r6
Shouldn't it at least be 'It's some *more* UFO nonsense'? or how about 'ugh more ufo nonsense from that same account that's been littering these threads with its detritus lately' ?
I'm intrigued by the timelessness & indefinite nature of the category label you selected to lump these into.
Have you considered that they might be part of the CIA? Seems awfully suspicious that they keep posting the same messages, trying to keep people from seeing the truth for themselves...
Speaking of potentially underhanded metagames in this space:
https://www.astralcodexten.com/p/open-thread-424/comment/225258377
Hadn't considered; I just checked though and the user shows a history of substantive comments - - elsewhere.
Appreciating that, and in fairness I might be having excessively lofty expectations that the audience here might resonate with the topic. I can only leave it to the author to weigh in, but I can share that generally the authors who write in this space rely on the readers' coming in with a fair amount of tacit knowledge on the history of the topic (typically the Majestic documents in particular). Otherwise the articles would be untenably long and repetitive of content available elsewhere.
I notice you still didn't mention what the whole Majestic thing is about, just "the people fomenting conspiracy theories rely on their readers being equally obsessed nutjobs".
The Fortean Times, bless their cotton socks, used to cover this and related topics heavily in the 90s/00s. For the AI précis, which is as good as anything:
"The Allegation: MJ-12 is alleged to be a top-secret committee of scientists, military leaders, and government officials formed by President Truman in 1947 to manage the recovery and investigation of alien spacecraft (specifically the Roswell incident).
Coverage: Fortean Times has explored the controversy surrounding the authenticity of the MJ-12 documents, which many researchers consider to be forged, while others (like the late Stanton Friedman) argued for their validity.
Context: The topic often appears alongside discussions of UFO crash retrievals, government secrecy, and the "all-star" cast of members supposed to be involved, including high-ranking intelligence officials"
So it's basically "little green men are real, Area 51 and Roswell are real, the government has flying saucers and is engaged in massive coverups".
You know, the usual UFOlogy from the 50s onwards.
By the way, funny you mention the Forteans. It's an interesting movement, yet the first time I saw them mentioned in any official context just happened to be earlier today:
https://www.cia.gov/readingroom/docs/CIA-RDP81R00560R000100030029-8.pdf
"Perhaps that'll take care of the Forteans for a while."
That's Howard P. 'Bob' Robertson of 1953 Robertson Panel fame, CalTech professor, & Alsos mission veteran. 'Chad' is H. Marshall Chadwell, head of CIA's Office Scientific Intelligence of the time, which asked for the panel to convene on the UFO topic and who responded with the typed-up report of the Robertson panel:
https://www.cia.gov/readingroom/docs/CIA-RDP81R00560R000100030026-1.pdf
The main practical outcome of the panel (pdf pp.23-25) was to recommend an 'education campaign' for "debunking" (scare quotes theirs) the UFO / UAP topic. The Don Marquis among those proposed (p.25) to enable the campaign was an expert academic on the psychology of propaganda (Needell, A.A. (1993), “Truth Is Our Weapon”: Project TROY, Political Warfare, and Government-Academic Relations in the National Security State*. Diplomatic History, 17: 399-420. https://doi.org/10.1111/j.1467-7709.1993.tb00588.x )
We happen to live in the timeline stemming downstream from that panel recommendation.
Those are all a fair summary. Replying to another , here's Gemini's summary:
https://astralcodexten.substack.com/p/open-hidden-open-thread-4235?utm_campaign=comment-list-share-cta&utm_medium=web&comments=true&commentId=223349883
In my experience, most of the readers here are strong skeptics of conspiracy theories and pretty biased toward the dominant narratives of mainstream media reporting, even when there's very compelling counter-information. Examples: COVID-19 lab leak theory, hesitancy/ questions about the COVID mRNA vaccine, Epstein in general.
I would expect them to dismiss most UFO stuff as a matter of course.
For me, you'd need good evidence to convince me. Not just a few people making claims. And it's gotten harder to convince me as the ease of faking evidence has increased...because people say all sorts of things, and whether it's fame or notoriety is in the perception of the viewer.
There is plenty of evidence out there, but you need the desire to find it and read it. "UFO's and Government" is a sober start. Also check out any Gary Nolan interview.
UFO believers still don't seem to grasp what level of evidence the skeptics are asking for. Not another trove of documents, no more sworn affidavits, no more appeals to credentials. These are entirely meaningless before the magnitude of the claims.
If the claim is that there are bodies, produce the bodies and let the world's scientists examine them. If there are crashed ships, produce superior materials unknown to the scientific literature, or demonstrate their reverse-engineered death rays. If there are reptiloids governing us, then, i don't know, shoot one in the leg and let their green blood flow on national tv. That is the level of evidence that would interest me.
I'm convincible. But I'm convincible on the object level and with some effort on the part of the person who's trying to convince me, not via tactics like this:
1. Link long post that assumes lots of prior knowledge and doesn't make it clear to an outsider what the important claims are
2. Instead of accompanying the link with useful background, just hint that it is Important Forbidden Knowledge:
> Beware, some ACX readers believe that the UFO topic is "stupid". And since you know that no topic has ever been stigmatized that turned out to bear merit, then there is no need for you to click-through. (You certainly won't benefit from doing so).
3. When met with incomprehension, refuse to elaborate.
Yeah I'm afraid the effort will have to come through you. The info is out there. (The post was very opaque.)
I'm with you, the original comment and follow ups weren't ideal for getting people to engage with the topic.
I commented more to contradict @Blocked Epistemology's assumption that ACX folk *specifically* would resonate with the UFO topic.
I should have added that my opinion was also based on rarely seeing the topic come up as a major point of discussion!
@vaclav
1. People on here link to technical posts all the time that necessarily assume some prior knowledge. Anything about MCP for example. (What is huggingface? What is an LLM? What is a transformer? Why is it called a transformer?...and so forth).
2. Was in reference to https://astralcodexten.substack.com/p/open-thread-423?utm_campaign=comment-list-share-cta&utm_medium=web&comments=true&commentId=222231227 .
3. I refused to elaborate? Do you mean to Labelle's comment?
--
@Christina the StoryGirl
>weren't ideal
Perhaps not. But I also don't know how many people go and start to inform themselves to a basic extent on the topic after my original comment. Maybe as a result when I comment in the future there will be different receptivity as a result.
>to contradict
Indeed I stand contradicted.
>rarely seeing the topic come up
Yes. Wish I had a media comms specialist who could explain to me what factors induce some audiences to engage with certain topics but not others. Then I could pick my forums with better efficiency, or at least with reduced push-back.
"coming in with a fair amount of tacit knowledge on the history of the topic"
I think you mean semantic or explicit knowledge? Tacit knowledge on the history of a topic would be quite a thing to have, presumably only obtainable from having been personally involved in a large portion that history! (Sorry, I don't mean to critique your grammar, just clarifying who you think the audience for this article is.)
You're correct., or 'background knowledge'. I'll leave the error in so people aren't confused by the comment.
Can anyone explain to me why anyone would *want* to use an LLM-based AI for mass surveillance? The use cases I have seen discussed starts with the government collecting (by licit or illegal means) many terabytes of raw data, for example cell phone locations. Then it is imagined that the AI does some magic and that a dossier gets created showing that Gavin Newsome and George Soros met in Fresno at 8pm on March 3 at 3pm. But if you want to extract the Gavin Newsome info from those terabytes you use a good old fashioned algorithm to do so. Using an AI would be too slow and expensive. And if, for some obscure reason, you wanted to train an AI on all those terabytes of data, you would start with a tabula rasa and spend several months and billions of dollars. You wouldn't start with Claude. Claude was already trained on its own terabytes of data for months and $billions. It doesn't have room to learn terabytes more.
What am I missing here?
I found this Ezra Klein interview with Dean Ball very enlightening on this topic:
https://www.nytimes.com/2026/03/06/opinion/ezra-klein-podcast-dean-ball.html
Ball was a former adviser to the Trump on AI issues. Excerpt:
=======================================================
Surveillance” is the collection or acquisition of private information, but that doesn’t include commercially available information. So if you buy something, if you buy a data set of some kind and then you analyze it, that’s not necessarily surveillance under the law.
....
The reason that no one really analyzes it in the government is not so much that they can’t acquire it and do so. It’s because they don’t have the personnel. They don’t have millions and millions of people to figure out what the average person is up to.
The problem with A.I. is that A.I. gives them that infinitely scalable work force. Thus, every law can be enforced to the letter with perfect surveillance over everything. And that’s a scary future.
NYT: "Homeland Security Wants Social Media Sites to Expose Anti-ICE Accounts"
https://www.nytimes.com/2026/02/13/technology/dhs-anti-ice-social-media.html
Even with current-level AI, it's quite feasible to just scrape all social media then say "hey Claude, analyze all data posted on reddit by acab_cop_hater, cross-reference with employment and social media data and identify their real identity, then draft a letter to their boss highlighting their problematic social media comments and initiate a mortgage fraud charge via the DoJ API."
Or, "given this high-resolution photo of this anti-ICE protest, use the ClearView API to identify all faces, determine the identity of all protesters, and set up an alert system that will notify us if they ever apply for a local, state, or federal job, or are ever subject to background search for private sector employment"
I agree that these are problematic and realistic uses of LLM-based AIs. My only quibble is that they are not examples of mass surveillance by DoD.
ICE is going to honkers's houses, getting out of the ICE vehicle, and waving at the honkers. Someone is clearly doing something (running plates, maybe?) Hilarious when the honkers call 911 -- for someone getting out and waving to you in your driveway??!?
"Identify all 2,000 people in this photo", or "cross-reference the stingray data from all 50,000 people at this protest" is not mass surveilance?
It IS mass surveillance, but as I expressed in my original post, I am skeptical that Claude or any off-the-shelf AI would be particularly good at it.
It probably has nothing to do with how you or I use it, LLMs are very good at specific search functionality which military applications might need for whatever they cannot cover with basic SIGINT
More effeciency around what they already do. We already have tremendous capacity to do analysis. OpenAI and Anthropic are lowering the cost per query so to speak. So do more with fewer gpus.
The use-case that seems intuitive to me is using it as a very sophisticated filtering mechanism. When you have a ridiculous amount of data, you can't reliably find the relevant stuff with blunt tools like keyword search, but it's impossible to get human eyes on all of it (let alone the eyes of a single person who can integrate information scattered haphazardly across the dataset). With an LLM, you're (potentially) getting something approaching human-level ability to notice what is relevant, connect it to the other pieces of the puzzle, and present the higher-ups with a summary and a pointer to the relevant pieces of raw data, but with vastly superhuman speed.
I think maybe the argument is that you can use AI to piece together a narrative from circumstantial evidence. Like if you fed a bunch of info as to what a jihadist was doing for a week, the LLM can be like "yes, this behavior is consistent with someone trying to make sarin gas."
I commented that, presumably the data would be part of the training regimen, and therefore the context window limitation doesn't matter. But when you put it like that, I realize my comment was dumb and I retract my statement. The data, will indeed, probably be fed into the AI during inference.
A more practical use, I suspect, would be to have the LLM do some bulk intermediate steps so that the human analysts can spend their attention on the bits that are harder or impossible to get an LLM to do well.
For example, you might have a million intercepted phone conversation recordings. An LLM could conceivably be tasked to transcribe the recording, summarize the transcript, and tag it with some subject metadata about the contents of the conversation. I expect very little training would be needed for this, since the first two parts especially are very similar to existing LLM-based features in video conferencing and medical transcription software. The last may take some specialized training but seems like it would be in an LLM's wheelhouse to do so in a way that is directionally correct enough to be useful.
The human analysts would then run database queries to find patterns in the tags, identify conversations they want to analyze, and then read the summaries or the full transcripts depending on how much scrutiny they want to apply.
I expect they're already doing something similar to foreign surveillance intercepts and the like, but with pre-LLM speech-tp-text tools and keyword searches instead of summaries and subject tags.
Since this is pretty much precisely how AI is currently being used in mainstream social research (think oceans of text-based raw qualitative data) even as we speak, I think this is a pretty reasonable assessment.
NSA was well beyond that in 1990.
> The human analysts would then run database queries to find patterns in the tags, identify conversations they want to analyze, and then read the summaries or the full transcripts depending on how much scrutiny they want to apply.
I would probably automate this part, too. Humans need time to think and write notes and remember what else they wanted to check... the AI just keeps going.
Yes. That makes sense. Transcription and classification of intercepted audio. A huge number of things to be analyzed, none of them requiring a large context window. Then follow up with more traditional software for matching and finding needles in haystacks.
I'd expect them to be applied to analyzing the content of texts, emails, transcripts of conversations, etc., not location information.
And they want to use an AI for the analysis because ...?
… there are too many messages to read them all?
… natural language processing is now a solved problem, so why reinvent the solution?
… it's better than filtering for keywords?
… what the fuck else would you use?
Yep. Same scenario as Erica raised. Makes sense to me now.
I wondered about that too.
I think we've seen enough in Iran that I can make a prediction for how it ends:
1) gradual air war drawdown within the next few weeks
2) weakened regime (probably but not certainly continuous with the current one on paper but with weakened ability to repress the people and far less weapons to threaten the west with)
3) a net overall improvement for both America and the people of Iran
4) enough problems that people who started out opposed will have something to point to and not feel forced to change their minds
I don't buy "net improvement for the people of Iran". The regime's ability to deliver *centralized* repression will be decreased, but repression is experienced locally and the local machinery of oppression will remain. Under essentially the same managers.
Except now a lot of them will be really pissed off, looking for payback against the people who just attacked their country but not being able to reach those people. And they won't be concerned by public image abroad, like the centralized regime sometimes was. And with nobody manning the "off" switch.
Military strikes on repressive military groups can be targetted. (Not discussing repressive clerics or culture, both of which can't be easily targetted). Given this, American strikes could conceivably enhance the "off" switch of "military involvement siding with the non-repressive civilians"
I'm concerned about a Libya-style fragile-state outcome. Obama called Libya the biggest mistake of his presidency, because the coalition knocked out the "bad guys" without any plan for what came next.
Obama's not being truthful about what the reasons for Libya were, if he's still talking "bad guys." I therefore cast significant doubt on his regrets (which, yes, are separate from lying his ass off about the intervention's reasons, but still! Liar should not be trusted!)
I didn't mean to claim Obama said that. But sure, I'll take the bait. What do you think Obama's true reasons were?
Maintaining Pax American via the petrodollar. It was specifically an intervention asked for by France... Qadaffi was talking a pan-African currency.
Seems dubious. https://africacheck.org/fact-checks/meta-programme-fact-checks/did-gaddafi-plan-introduce-african-currency-backed-gold
People will believe anything as soon as you offer the thinnest appeal to self-interest, even if it makes little sense on its own terms. Why would the US do airstrikes to defend the use of France's currency? Or else, why would France request airstrikes for the sake of the American petrodollar? Why did the airstrikes coincidentally happen when Libya was in the middle of a civil war? If the US cares so much about its reserve currency status, why does it wield sanctions so often in a way that creates incentives to adopt alternative currencies? If the US cares so much about its reserve currency, why is Trump trying to reduce the trade deficit that's an ~inevitable byproduct of reserve status?
What's your agenda Zanni?
My agenda? I ain't got no agenda, sure nuff. I do keep my ears open, and the folks I'm listening to have more access than "africa check" does.
The United States wants its vassal state, France, to retain control over "the francophone african states", rather than ceding them to Russia or China (I think those are the only big non-West players fishing in Africa). This is basic geopolitics.
France benefits from the petrodollar and Pax Americana (specifically, in keeping shipping open and removing pirates from the open seas). One can argue how much France actually understands this, however.
Airstrikes during a civil war? That seems like an optimal time to do them, particularly if the civil war was... shall we say, western inspired (I have no knowledge here, simply laying out a suspicion).
The United States very badly miscalculated in its sanctions on Russia. Other than that, there is no "so often" -- we've left SWIFT and the other banking institutions as a gun on the mantle, ready to be pulled when needed.
Nuland and the neocons broke Pax Americana, so I wouldn't be so certain that the petrodollar remains the reserve currency in the world. Trump's certainly not banking on America being the unipolar power that it's been -- he's putting down chips on "we need to do this to ensure there IS an America in a decade." Hence Canada, Greenland, Venezuela (draw a circle, and then write Golden Dome).
How exactly is this an improvement for America?
If the answer is, “we won’t have to spend as much money defending Israel,” I know a much easier way to do that.
The US is already winding down the (surprisingly small) amount of support it gives Israel (at Israel's request, since they don't want the dependence either). Which was always more of a subsidy to keep defense contractors in business (useful since you don't want to have to suddenly spin them up from zero if you suddenly need them) than anything else.
US has pre-existing commitments to fork money (materiel) over to Israel in the event of an attack. Gets written into every budget, but is zero without an October Attack.
Israel has been the top recipient of US foreign aid for many years, several billion dollars a year. Is that "suprisingly small"?
US foreign aid (not including military) is about 50-70 billion/year, (foreign military aid varies wildly but was about 10-20 billion/year total before the Ukraine spike, not counting things like nato-suppirting bases abroad or deployments in Afghanistan). Military aid to Israel is about 3.1 billion a year (and as mentioned is being wound down to zero over the next few years- there was a brief post-2023 war spike but it never approached e.g. Ukraine levels). It's just not much of an item. People love talking about it because of, uh, (((reasons))), but it was always a small piece of the foreign aid budget which is itself a rounding error in the federal budget.
I think the complaint is more that US middle east policy is dictated by Israel rather than us sending Israel a bunch of aid.
I don't understand why all those Nobel Prize winners had the brilliant idea to build their state in the world's most quarrelsome region, right next to the people they have all their historical grievances with. Even I, as a feeble non-Nobel intellect, can see that Moon Zionism is clearly superior.
It wasn't obvious in the 19th century that Arabs would be the world's most quarrelsome people in the 21st century (also, could've been much worse - imagine if they'd somehow gotten the land for it somewhere in Europe, the Germans would've just killed them all. And if they had gone with the Uganda plan... It's not like Africa is known for being peaceful. And FDR vetoed the Alaska plan. Basically no good choices)
Since their founding, we've given Israel > $300B, ~$250B of which was military assistance. So I guess if you thought we'd given them trillions, it'd be a surprise?
Alternatively,
3) for both America and the people of Iran, things ultimately get worse on net overall
4) enough localised things being better that people who started out in favour will have something to point to and not feel forced to change their minds
My general disappointment with the state of public discourse in the post-truth era leads me to suggest that in reality the camps from your (4) and my (4) will exist simultaneously, and it will take significant work to decide which reality we are actually in for those who still care.
4) is an interesting alternative, but I think it is far less likely than you think. Iran has a co-ed dance scene, it's just underground, and for the middleclass urbanites. I'm pretty sure a lot of people who want to say "Iran is crazy" don't realize they've actually legalized prostitution.
It's good the regime is being dismantled in real time. But whether or not what follows is worse or better, no one can say. Meanwhile there is a huge human cost to bombing a country that will be glossed over by the hawks. Net good, but: be cognizant of the lives being snuffed out.
It is striking the degree to which many online American discussions of this and other wars seem completely oblivious about the fact that human beings (other than Americans) die in it. Not explicitly dismissing the value of Foreign lives, not accepting it as a fair price to pay for a greater good, just... not even recognizing it as a thing that is happening.
Grappling with costs is a huge cognitive load. It's easier to play Geopolitical Simulator when those are abstract figures.
The War Crimes were occurring before the bombing started. Water War, led to protests. So, the "huge Human Cost" is perhaps LESS than leaving Iran alone (and letting the Israelis play water games in Afghanistan).
If the successor is worse, the current regime being dismantled isn't good.
That's incredibly unlikely though. There's currently basically zero regimes that are worse, including places like Libya (north Korea is borderline).
Iran is arguably worse.
Iran is worse than... Iran?
The Iranian regime now is a worse than ...
How do you figure? What do you think is so bad about Iran, the country that thought about getting nuclear weapons, and then decided "nope!" (around the time of the Axis of Evil.) Bibi's been saying Iran's very close for 30 years, for god's sake.
This kind of thing often ends up with a worse government replacing a bad government. And anyone anti-US will have a lot of popular support. (There are lots of ways for a government to be bad.)
It kind of makes sense. "Look what they did to us. I'll stop it from happening again, and MAKE THEM PAY!" That seems like a good sales pitch for a new regime.
Yeah, I agree with this. My expectations are something like 20% end up net worse, 30% roughly enough improvement to balance out the damage the direct strikes cause (this is lower than it would otherwise be because of how consistently the regime has supported civil wars all over the world, especially in Syria and Yemen), and 50% net improvement that's worth the cost. I don't want to gloss over the human cost here.
What are the odds that Iran won't turn into another Libya hell hole?
Unlike Libya or Afghanistan, Iran isn't a pre-state tribal society. It is in fact one of the oldest states with continuous existence, over 2500 years. They won't collapse into a squabbling bag of warlord-operated territories.
What I am afraid of is authoritarian state on steroids. The worst element of the IRGC surviving and unleashing endless hell on the population. They may have run out of ballistic missiles (or almost), but they certainly have a lot of ordinary bullets.
Hopefully it won't come to that.
> They won't collapse into a squabbling bag of warlord-operated territories.
Even if Iran is reduced to rubble? I'd expect even European countries to turn to savagery if their entire government and half of their country was nuked. People act differently under desperate circumstances.
Much of Europe was reduced to rubble in 1945, just look up photos from Poland or Germany. Old nations tend to re-create new government structures quite fast, because the know-how is there.
That said, Iran won't get "reduced to rubble". It is mostly the regime infrastructure that is targeted. The US and Israel don't even have enough conventional munition to level megacities like Tehran, they would literally have to nuke them. Dumb ammo is mostly gone from Western arsenals and smart ammo is too expensive per unit to be intentionally wasted on civilian buildings with no military value.
(That does not exclude mistakes, which always happen.)
Tehran doesn't need to be leveled, due to ongoing warcrimes.
I sometimes wonder if using nukes in a situation like this would be that much worse for optics compared to leveling a place using conventional munitions, like Israel did with Gaza. Yes, a nuke is a nuke, but... I was under the impression that the newer versions were more "clean", relatively speaking. Nobody is going to start a nuclear war with the US over this.
That would still be an improvement over the IRGC! They killed more people in a single weekend than died in Libya in the last 16 years combined (and that's just domestically, not even looking at things like the IRGC-backed Yemeni or Syrian civil wars).
I think Trump fumbled when he led with his mouth instead of with strikes. Giving the Mullahs time to massacre protestors was a bad move. It looks especially bad now that we're actively engaged in the conflict. What percentage of people who were interested in regime change were killed? How many remain, and how many are not cowed into silence by the killings?
Like almost everything this administration does, this was not handled with prudence or care. A maximal show of force should not distract us from that.
The massacre of the protestors allowed identification of sympathetic military battalions. Do you want to murder all the "good troops", and leave just barehanded civilians?
Yeah. I think realistically the protests took everyone by surprise and it took time to assemble forces, but a mix of talking slower and possible early partial strikes could have made this a lot better :/
Not everyone. the people causing the war crimes were not surprised. Alternatively, the protests were a CIA thing* that just caught fire bigger than planned...
*it's possible to have locals who want to protest, as well.
I think the toys were just being moved to proper battlefield positions and Israel was ready - after the slaughter of civilians. It was inevitable.
Okay, I'll post my predictions:
1) A lot of innocent people will be slaughtered, most of them by the US, far more than the Iranian government was killing.
2) The Islamic Republic will survive. I'm very confident on this part, because history has repeatedly shown that you can't install a new government from the air and there's no public will in the US to commit anywhere near the number of ground troops they'd require. (and even if they did, it'd end up looking like Afghanistan on steroids.)
3) Oil will get significantly more expensive, with knock-on effects on economies on the world, in aggressor and neutral countries alike.
4) The war continues for much longer than Trump or the US desire, because Iran won't settle for anything less than the gulf emirates evicting US bases *and* quite understandably distrusts a country which withdrew from the prior peace agreement, assassinated their head of state, and used peace negotiations as cover for a surprise attack.
2. Serbian regime in 2000 did, in fact, fall after a bombing campaign. The situation was somewhat similar, plenty of people hated Milošević.
3) Will get? Gas station prices already jumped up 20% where I live (central Europe) since the attacks started .
1) This is awfully vague. How many is "a lot"? A Google search indicates Iran executed 1000 to 1500+ people in 2025, which is a "significant increase in the use of the death penalty". Is your prediction that the US will kill more than 1000 civilians? It seems likely, given war, even with precision strikes.
3) Iran produces oil, but is #8 in the world (https://en.wikipedia.org/wiki/List_of_countries_by_oil_extraction) before any damage to oil production. Again, it ought to have some effect on oil prices, but "significantly" is too subjective to be proven wrong. Plus, we have no baseline for what oil prices would have been otherwise, and many things can influence oil prices.
It may be the #8 producer, but it can close the Straight of Hormuz, and significantly affect the Bab al Mandab via the Houthis. ~35% of global seaborn oil flows through those two chokepoints.
During peacetime, yes, they can certainly close it. But during wartime, I do not doubt that the US Navy can keep it open. I have already heard some things about ensuring or insuring passage, and Iran's navy was destroyed Saturday, as I understand it.
Even if the US Navy can ensure safe passage for 95% of ships transiting the straits, even a 5% risk of constructive total loss can completely upend the insurance situation. A Navy-less Iran can still credibly threaten shipping directly off its coast with cheap Shaheds that are very difficult to suppress. If insurance declares the route too risky to cover, then it's effectively closed, even if there's a good chance any given ship would make it through unscathed.
Could Iran attack a carrier in the straits, with a full normal escort complement? If not, could not a tanker, or more than one, take the role of carrier, and thus be fully protected? If so...why are these carrier groups so expensive?
I'm no expert, nor even adjacent, on military matters, but it seems sound to me.
The US has talked about providing some sort of insurance backstop.
On the other hand I would want to think about the poor crews of these tankers. A 95% chance of survival with a US-backed insurance policy might be acceptable to the oil company actuary but it's not going to be acceptable to the poor sod who has to be on board. If you tell me to sail half a million tons of crude oil through Hormuz right now then whoops I guess my oil tanker suddenly has mechanical problems which will require lengthy repairs.
Shaheds are not that difficult to suppress, if you ask the Ukrainians for expert knowledge and have as rich pockets as Arabs and Americans have.
Remember that original Shaheds are already obsolete in the Russo-Ukrainian war, precisely because they were relatively easy to suppress. The Russians are now using newer, more advanced models (Geran-2) that they produce themselves. Iran does not produce this new generation.
I agree that 1-3 are most likely.
1. Even more people will die from the breakdown of basic services and order than from direct Israel/US bombings.
2. Islamism is still widely popular. Maybe the next leader will not be an ayatollah but he will still be a hard line islamist. Or maybe Iran will break down in a sort of feudal order with local military leaders empowered by the destruction of the national leadership.
3. Oil has already become more expensive. Plus, the worst for iranians has already happened. It is very likely that the iranians will try to block the Strait of Hormuz as there are no downsides to escalating now.
4. The aggressors keep killing the iranian leadership so there is no option for peace talks at this moment as there is no one who can lead the peace talks.
For context:
https://manifold.markets/GP/will-the-price-of-oil-reach-100-per?r=VkRF
https://www.metaculus.com/questions/38770/will-the-iranian-government-lose-power-before-2027/
I will create some more to help you work through those predictions
If ever you are still reading: https://manifold.markets/VDE/how-many-confirmed-deaths-will-be-a
If you do hold those views, please enter your predictions here!
https://manifold.markets/VDE/how-many-confirmed-deaths-will-be-a
If you can review this question that would be great as well: https://www.metaculus.com/questions/42531/confirmed-deaths-in-the-iran-war-as-of-march-31-2026/
https://manifold.markets/SaviorofPlant/will-irans-regime-fall-in-2026
https://www.metaculus.com/questions/7770/end-of-islamic-republic-in-iran/
For 1), which number of people killed by the Iranian government is your baseline? The death tolls for January range from 3,000 to 36,000 according to wikipedia.
https://en.wikipedia.org/wiki/2026_Iran_massacres
Reports on casualties in the current war are at around 1,000 at the moment, including both military and civil victims, so it's relevant which number you compare to.
I said "will be" for a reason, and I expect a long enough war that even the high counts become dwarfed by the conflict's end.
For calibration purposes, Iraq, Afghanistan, Syria, and Russia-Ukraine are all comfortably into six figures.
Thanks, that is more concrete.
Do we have estimates for the civilians currently killed vs previously in crackdowns?
> Iran won't settle for anything less than the gulf emirates evicting US bases
That seems WAY beyond their capacity.
This may be true.
However, a country having a war aim it can't achieve militarily and absolutely no faith in the negotiators on the opposite side is a recipe for a prolonged conflict.
Yeah, the thing counting against that is Trump's tendency to lose interest in things (other than tariffs) and unwillingness to stick to anything difficult.
This isn't an exclusively Trump's war.
MbS, Netanyahu, the Thanis etc. have some interest in seeing a resolution. Iran didn't help themselves by lashing out blindly on their neighbours from Azerbaijan down to Oman.
MBS doesn't like the Iranian regime, but I don't think he was launching attacks at the opening either.
What "Thanis"?
What does “weakened ability to repress the people” look like? And you you think we can form a view on the impact on Iran’s nuclear program?
I think it will (probably) be harder for the regime to shoot thirty thousand people in a weekend again. Possibly in a patchwork way (e.g. kurds or people in some provinces will be armed, and in other places the basij will be destroyed or unable to coordinate). And I think they'll be a bit more reluctant to do massacres before compromise in the future.
For the nuclear program I think it's harder to tell - probably some setbacks but way too early to tell how much. The ballistic missile program seems severely degraded, and afaict Iran was planning to rely on it to protect them while they raced to build nukes.
(Overall I see like, 5-6 different ways in which things might improve. When I say I think this will probably be a net improvement, I think the modal scenario is something like 3-4 of those pan out, while 1-2 new bad things that I didn't list also start happening. This is pretty unsatisfactory as a prediction that can actually be graded in hindsight, but it's a broad picture of my expectations).
What are the objective criteria for “improvement”? Iran did not threaten the US in any way before this war. And it won’t threaten after. Talking about real threat not talking heads on TV.
Trying to murder Trump doesn't count as a threat? Seriously? He's not the only guy they've got on the "to be murdered" list, either (which is a way of saying "destroying military commanders is destroying operational capabilities")
It depends what you count as a threat - they've certainly been a geopolitical adversary including training and arming terrorist groups around the world and shutting down the red sea for trade last year. If you're a more hard-line isolationist that's an argument you can make (not getting into the more general argument about isolationism now), but it's at least no worse an investment than e.g. supporting Ukraine against Russia (which is far less threatening to America than Iran was).
You haven’t answered the question. Iran supporting Hezbollah or whoever else is/was bad for Israel. I don’t see how it had threatened the US.
I don’t think Middle East will be safer for the US after this war. Despite Iran getting pummeled.
I think the Houthis were a bigger problem for America, since they'd been disrupting shipping.
Hezbollah does plenty of anti-american terrorism including working with cartels in Latin America. And Iran has done plenty of directly anti-american stuff (like arming anti-american terrorism in places like Iraq or Yemen, or assassinating American scientists and trying to assassinate Trump). And an Iran with a huge (potentially nuclear armed? ICBM stockpile wouldn't stop at nuking Israel.
Look, if you're an isolationist in general that's fine like I said. But Iran is at least as much of an American adversary as Russia or isis (or China for that matter), which the US invested significantly more resources into countering.
Should be easy to quantify then. Hezbollah had killed X American soldiers in the past 10 years. In the next 10 years it will be X/5. Or something like that.
You are just hand waving. Without giving any objective criteria.
Note that if you only count successful killings, and ignore the failed ones, you underestimate the costs, because preventing the unsuccessful attempts also costed you something.
Even worse, it would lead to logic like: because we were careful about them, they failed to kill anyone, therefore they are harmless, therefore we no longer need to be careful.
(This is a general note, not specifically on this topic. But I noticed how you deflected "assassinating American scientists and trying to assassinate Trump" by asking how many soldiers were killed.)
I argued with like 50 people on the Anthropic thread. One of the arguments people were making was "we're only talking about DoD/DoW use of AI for mass surveillance, no one is talking about police doing their job"
Well, here is an older report that says that Anthropic literally will not allow any law enforcement to use their product. https://www.semafor.com/article/09/17/2025/anthropic-irks-white-house-with-limits-on-models-uswhite-house-with-limits-on-models-use
Again, the implicit argument people are making is that places like NYC have *just the right amount of enforcement* (I don't see many calling for ripping down the NYPDs camera system, which allows it to have one of the highest clearance rates in the country - and allows NYC to have some of the lowest crime rates in the country, especially given its demographic makeup), but making it any more efficient with AI would be immoral and unethical. It's a confused and nonsensical position.
The countries that use mass surveillance use a system that's nothing like Claude looking at a bunch of feeds and then making a bunch of accusations. You have cameras and ML, but it's not an LLM and it's not making any decisions.
You explicitly want this system to improve speed and massively increase the number of reports, this will inevitably lead to whatever human you put in the loop being a reverse centaur, just signing off on all the LLM's decisions, and only being there so you have someone to blame once the LLM fucks up.
Add to that that the courts simply can't handle a massive increase in cases. People would just dispute all the tickets/charges for littering or whatever en masse because they know the courts don't have the capacity to prosecute them all.
Improving speed behind a bottleneck doesn't increase the flow rate through the bottleneck and in many cases actually reduces it.
You have to explicitly make the entire system much more illiberal if you want to achieve the same speed and efficiency in prosecuting petty crime as those already illiberal places you're no doubt using as a benchmark(like China or Singapore). Just adding cameras and machine learning is not enough.
When it comes to courts and charges and clogging the system...
Just today I heard a (Czech) podcast with a realtor who sold apartments in Dubai. He commented on the local surveillance culture: "If you do as little as showing someone your middle finger in public, you will receive a fine within 15 minutes on your phone."
The punishment mechanism can be automated as well, though it is possible that the very distinct US judicial tradition would not allow for that.
>The punishment mechanism can be automated as well, though it is possible that the very distinct US judicial tradition would not allow for that.
It absolutely can. If you don't care about small things like right to appeal.
There's some argument to be made that the societal gains from increased public order would trump the losses from injust punishments, especially if we're only talking mild fines, but the US system is absolutely not set up to do this.
It would take a massive societal and judicial reform to get it there, my point was that just letting an LLM sift through camera feeds would be of little benefit and might just make things worse.
I am not endorsing that way of punishing people, I am just pointing out that this is possible.
The US criminal justice system is already quite heavy tilted toward plea bargaining, to the degree that the original jury system only applies in 1-2 per cent of all criminal cases (IIRC). This looks like an "optimization" of a similar sort, and this happened gradually, without a Day D which would mark a major reform.
I think my concern is specifically that the people in power will abuse this new system to attack their political enemies. With an AI to scale it, they could (for example) make a rule that anyone who probably voted against them in the previous election gets extra-strict enforcement for littering and jaywalking. This might not be difficult to notice but it would be very difficult to prove; the government would say "look, they totally did it, we have all this evidence!"
Trump is already doing some stuff like this on the state level; I searched for "Trump cuts aid to blue states" and found a bunch of articles. If we gave him the ability to attack blue voters individually, I think he would use it. And, to be clear, I wouldn't trust the Democrats with power like this either.
Downthread, you've argued: "It would be very good for the country to move toward fewer rules that are more consistently enforced." I agree! But I'm worried that this AI stuff will take us in the opposite direction from that.
Adding a second line of thought in a second comment for clarity:
There's a difference between 'Surveillance cameras in public spaces have been good for stopping crime, therefore we should add more cameras' and 'Surveillance cameras in public spaces have been good for stopping crime, therefore we should have an AI processing all the camera feeds and acting as an oracle for police.'
AFAIK, police don't randomly review security cameras all day, randomly searching for anything that might be a crime. Instead, when a crime gets reported, they look at the recordings from the area to see what happened, or look at current feeds once they have a suspect they're looking for.
More cameras would give them more coverage to do those activities more effectively.
But adding an agentic AI that can process all the feeds at once and put together queries, databases, etc., is a qualitatively different thing.
For example: an AI monitoring every feed can tag every litterer in the city, but the police don't have the manpower to prosecute every litterer in the city; nor every jaywalker, every speeder, every person who parks an inch too far from the curve, etc. Instead, they end up with a database of minor offenses for every citizen in the city, and can decide at their own discretion who to prosecute or not, can have a list of offenses they can hold over anyone they want to go after or who challenges the m or displeases them, etc.
Mayube yu think that's good too, maybe you think we'd reform things such that this capability was used responsibly and led to good outcomes.
But my point is that this is a qualitatively different set of capabilities than what modern security cameras grant the police. Therefore, 'cameras have been good, therefore AI will be good' is a non sequitur.
> For example: an AI monitoring every feed can tag every litterer in the city, but the police don't have the manpower to prosecute every litterer in the city; nor every jaywalker, every speeder, every person who parks an inch too far from the curve, etc. Instead, they end up with a database of minor offenses for every citizen in the city, and can decide at their own discretion who to prosecute or not, can have a list of offenses they can hold over anyone they want to go after or who challenges the m or displeases them, etc.
How hard can it be to go after every speeder/illegal parker? Lots of countries do this: There's a tired civilian cop that gets a new photograph every minute on his or her screen, validates that it looks what the description says it looks like, clicks the big green button, the owner of the car is sent a payment link to pay the fine. Sixty fines an hour, 450 fines every working day. If the car owner wants to dispute the fine, they are free to do so.
It's harder to do this with litterers/jaywalkers, because facial/gait recognition is much more subjective. Modern neural networks can be surprisingly good, but the human in the loop won't be able to confirm their guess, not unless the AI does something like: "Here's Ms. Doe leaving her house in a green jacket and sunglasses, photo courtesy of her Ring. Here's a receipt for an ice cream, paid be her card at 12:50. Here's a woman in a green jacket and sunglasses, same height, same hairstyle as Ms. Doe, eating an ice cream, throwing an ice cream wrapper into a bin and missing at 12:52. Here's her path through the city, courtesy of her phone, that places the phone in the vicinity of the bin at 12:52. Here's the list of all other people seen in the vicinity of the bin at 12:50-12:55, all of them have a negative match with the litterer."
The first problem with this approach is that now a cop needs at least five minutes per fine to issue it. 90 fines a day is still nothing to scoff at. The second one is that now that standard of evidence is lower. "Black BMW, license plate 2FAST4U" is a fact that is hard to challenge when captured by a speed camera. With Ms. Doe we have to learn to evaluate and accept this kind of circumstantial evidence that *excludes* other perpetrators, Holmes-style, instead of fingering her directly. And be we I mean the judge and the jury, because this *will* be challenged. If not by Ms. Doe, then by Ms. Chang or Ms. Washington.
I don't agree with you. The answer isn't to give up on crime enforcement, it is to actually fully enforce every law on the book.
"For example: an AI monitoring every feed can tag every litterer in the city, but the police don't have the manpower to prosecute every litterer in the city; nor every jaywalker, every speeder, every person who parks an inch too far from the curve, etc."
If there is a law on the books that you don't agree with then take it off the books. I agree that discretion is bad. Your response is to say "then we just shouldn't monitor crime". Mine is: no, we should definitely monitor crime, but use technology to enforce everything that is on the books. If you don't want something to be a crime, then take it off the books. In a just society, someone who litters, tags graffiti, etc gets a nice juicy fine in the mail the next day. And if you think that isn't fair then make graffiti or littering legal.
Your preferred system is that we should just have it be luck based who gets punished (a cop happens to be nearby)? That's crazy!
I don't think it's physically possible to enforce every crime, including minor crimes such as littering or jaywalking, without automating the enforcement along with detection; and detecting crimes while lacking the ability to enforce them is a waste of time and money.
Currently, the criminalization of e.g. jaywalking acts as a deterrent. On the one hand, when you jaywalk there's a small chance you'll be stopped by a cop; but on the other hand, jaywalking can serve as a factor in the prosecution and sentencing of other crimes. For example, if you get run over by a car, the driver would normally be found liable -- but if the traffic cam video shows you jaywalking, then his liability would be reduced, potentially down to zero. Due to these factors, some people who would normally choose to jaywalk abstain from doing so.
Merely automating enforcement won't get you very far, unless you fully commit. For example, you could stick a camera on every corner and automatically mail out tickets to jaywalkers. Some will choose to pay the tickets, but most will probably ignore them (this is what often happens with speeding cameras in the real world). The next step is to summon the alleged criminals to court, but in order for that to happen, your courts need to have the capacity to try all those cases, which they currently do not (they barely have enough capacity for speeding tickets). As the result, most jaywalking cases will remain unpunished.
The next step is to automate the courts as well, or bypass them altogether: if the jaywalking AI deems you guilty, then you're guilty, and refusal to pay the fee makes you increasingly more guilty until the police are dispatched to your house to put you in jail. This could work as an effective deterrent, assuming you put in enough effort in the initial phase of your crime-fighting campaign to stage some very public arrests. That said, the cops don't have enough manpower for a sustained effort like this ether, but perhaps you could optimize them by putting the AI in charge of dispatching cops as well.
And now, you live in a paradise where almost no one ever jaywalks; in fact, people are rarely leave their homes at all, for fear of being randomly thrown in jail by an unaccountable LLM with no due process and no hope of exoneration. Like I said... paradise !
You have a distinctive profile image, so I recognize you from some of your other comments.
Frankly, I think that most of your comments have been snarky, glib, and dismissive. Most of the people who you've responded to seem to share my opinion.
In contrast, I think that this most recent comment of yours was articulate, fully explained, and polite.
In case you found yourself putting substantially more time or effort into this comment than your other ones, I want to let you know that the difference is both noticeable and appreciated.
To be honest, when I post my comments I don't set myself a goal of being snarky vs. articulate; rather, I always strive to convey my honest opinion on topics that I find interesting. Sometimes I express agreement, sometimes I express disagreement, and sometimes I express strong disagreement indeed -- it all depends on the topic. That said, I'm only human, and as such I've got to admit that I am pleased by and thankful for your attention !
When I was in Singapore and Shanghai I certainly saw a lot of people walking around!
That's because Singapore is a small country that actually does have the resources to enforce all of its laws (with no AI automation); and Shanghai is a city in a country that is fully onboard with implementing the police state and all the joys that come with it. I don't know what the situation is like in Singapore, but in Shanghai, the average citizen is in constant danger from his own government -- much more so than the average American is from jaywalkers.
This is the same thing as the SAVE act: I'll support requiring ID to vote *after* we have a competent bureaucracy in place with a 10 year track records of making sure every citizen has valid ID or can get it in a few days with under an hour of effort. But what they're offering is to require an ID to vote *while* the bureaucracy is a complete mess.
Sure, I am absolutely in favor of reducing the number of laws and regulations, and the schedule of punishments, down to a level where I would be happy with 100% perfect enforcement. And after the legal system looked like that, I'd be happy to give the police the tools they need to reach 100% perfect enforcement.
But you're advocating doing the second step first. And I'm saying that will lead to horror and dysfunction.
I too wish we were already in utopia and could act accordingly. But I won't apologize for recognizing the world we're *actually* in, and acting accordingly for that world.
When you succeed in reforming the entire legal system into a perfect ideal that can support 100% enforcement, get back to me and I'll join your side.
Assume that 50% of Americans no longer trust when a Democrat gets elected. Do your priors change now? The ID - requirement can be as performative as the TSA (literally never passing inspection), and still be good for America.
No, you don't reward liars by giving them the thing they're lying to get. Especially not when it not only incentivizes, but also directly empowers them to keep lying more in the future.
As they say, the decision theory on this is straightforward.
So, it's your assertion that you actually know about the ballots, and have crystal clear evidence, contra the boxes carried out of Fulton County by the FBI? That seems rather speculative on your part. I wouldn't sit around and say "I know that election was fair" -- not when we have more than one party* that would rather bollox the election, on general principle, and more particularly because of one Donald Trump, existential threat to the Middle Kingdom.
*China. Russia.
One of the things AI is changing is that a whole bunch of processes were naturally bottle-necked by human attention, and they were calibrated to have that limiting factor, and now that factor is gone.
"Well, duh, fix those" okay but we have lots of things to fix. Breaking it to find out what we need to fix is going to cause lots of problems and we already have a big list of broken things already.
https://nitter.net/patio11/status/2024197539344375891
I'll say the same thing I said in that thread - surveilling public spaces with cameras, and surveilling private correspondences and actions through data mining, are not the same thing.
I don't actually know how much public surveillance NYC does and how much it actually helps with crime, I definitely don't naively trust anything the NYPD claims, if I did a deep dive I might want more or less than the current level, I don't have a strong prediction.
But that's really not what we're talking about here. I think it's bad for people and bad for the country if every action you ever take on or near a digital device in the privacy of your own home or place of business is being watched by the government. Adding an agentic intelligence that can actually process all of that information at once and act as an oracle for government agents makes it much worse.
Public surveillance sucks, but you were already being surveilled in public by the other people in public, you already had to steel yourself and put on a public face and be on your best behavior until you made it from one private space to the next.
Being under surveillance *always*, having *no such hting* as a private space *ever*, is bad.
Bad for the soul, bad for society and culture, bad for our relationship to the government.
The point of the Panopticon as a thought experiment is that the Panopticon is bad. When you are forced to build a policeman inside your own head and perform the actions of innocence and compliance for it at every moment of the day, you eliminate the space needed for your own individuality and humanity to flourish.
Trust me, you do not actually want to live in the Panopticon. It is not fun and not worth it.
Stop trying to build the Torment Nexus.
You are exactly who I was referring to! There is no evidence whatsoever that Anthropic is exclusively saying "don't use the tech to data mine everyone's digital devices" - the other article lays out fairly conclusively that they just don't want their tech to be used to uphold the law.
This is a classic position I see from a lot of Western leftists, when you don't agree with a law (or are uncomfortable with it for whatever reason), instead of changing it, just throw as many barriers in the way of being able to faithfully uphold it.
Anthropic's position is implicitly that police and related law enforcement are the only careers in the country which aren't allowed to have productivity increases through technology.
>There is no evidence whatsoever that Anthropic is exclusively saying "don't use the tech to data mine everyone's digital devices"
Who cares about 'exclusively'?
If one of the 5 things they are doing is horrible, I don't want them to do the horrible thing!
> instead of changing it, just throw as many barriers in the way of being able to faithfully uphold it.
First of all, I'd love to change the laws I disagree with! But I don't have unilateral power to change the law whenever I want!
There are plenty of activist groups trying to reform the actual law, and I support them. But given that they're not close to success on a wide variety of topics, of course I also support other efforts to mitigate the harm caused by bad laws. Why would I limit my efforts only to things that won't work?
Second of all: the legal system in the US has never been designed with the intent of enforcing all laws at all times. That's an insane thing to think.
Between jaywalking, littering, speeding, intellectual property, a 2600 page federal tax code, different buildings codes by states and county, byzantine contract laws for every different industry, and everything else, there's probably not a single adult citizen of the US that has never broken a single law or regulation.
Perfect enforcement would put everyone in jail, including the police.
Selective enforcement has always been the law of the land, and this has always meant that the police have discretionary power to wage cultural and political wars against groups they see as targets and enemies.
Giving them more power to notice small crimes and generate evidence doesn't lead to better enforcement, because enforcement was already limited by how much the police feel like enforcing and how much manpower and time they have to do it (plus court availability etc). All it does is give them more discretion about who to prosecute, and more leverage over anyone who protests.
You should read Robin Hanson on selective enforcement:
https://www.overcomingbias.com/p/why-weakly-enforced-ruleshtml
It would be very good for the country to move toward fewer rules that are more consistently enforced. In fact that is what separates a lot of developed countries from developing countries.
Yes, and Anthropic working with police does absolutely nothing to accomplish that goal.
Disagree completely. Part of enforcement is being able to detect crime in the first place! Technology definitely helps with that.
While I see your point in general, I don't see why the better solution isn't just "get another contractor to do that". There's lots of other AI companies that are basically almost as good (and will be better in six months than anthropic is now).
Can you expand on the point about NYCs demographic makeup?
Yup. If you just compare homicide rate for each ethnic group individually, NYC is near/at the lowest.
Out of 64 major counties:
Black homicide rate per 100k: Queens = 9.7 (2nd lowest in the country), NY County (Manhattan) = 13.0 (7th lowest), Kings County = 13.2 (9th lowest), Bronx = 16.3 (13th lowest)
Hispanic (only 48 have enough hispanic homicides): Queens = 3.7 (2nd lowest), Kings = 4.4 (8th lowest), Manhattan = 4.9 (10th lowest)
White (only 57 have enough white homicides): Manhattan = 0.7 (lowest), Queens = 0.9 (3rd lowest), Kings = 1.0 (4th lowest)
Still it is somewhat crazy that even in NYC, the White to Black difference exceeds 10:1.
> Again, the implicit argument people are making is that places like NYC have *just the right amount of enforcement* [...] but making it any more efficient with AI would be immoral and unethical. It's a confused and nonsensical position.
I disagree with this interpretation.
By analogy, this is like saying that if you support the US Military, you would want them to have the ability to kill anyone in the world instantly, without cost or consequence to themselves. But if they had this ability, would they kill the exact same set of people that they do now? Or would they also be tempted to do way worse things? Maybe the President would decide to kill every member of the Iranian government and the IRGC and anybody else the CIA said was bad. Maybe the threshold for targeting people would go down to a lower level of consequence. Maybe somebody would ask why we don't just kill everybody who looks at us funny.
Obviously AI doesn't let the NYPD vaporize shoplifters but it's still fair to wonder whether it might have some other impact on their behavior if ticketing people or surveiling people became much more trivial.
Also you're just completely ignoring the possibility that hallucinations at the scale of a 10M person city could add up to a lot of issues.
I think we just live in completely different universes if you think that NYC has even close to the optimal amount of punishment for crime. Anti-social behavior carries virtually no penalty. I could fare-evade, shoplift, litter, tag a building, etc. literally every day for the rest of the year and there would be 95%++ probability that I would not face a single consequence.
And that's NYC, with relatively high clearance rates (due to cameras!). The clearance rate in other U.S. cities is abysmal.
> I think we just live in completely different universes if you think that NYC has even close to the optimal amount of punishment for crime.
Which is not actually what I said. My point is that, one, giving somebody more capabilities doesn't just result in them doing exactly the same thing except more or better, and two, I have concerns about using unreliable systems in a criminal-justice context.
Law enforcement isn't just, like, one dial for "amount of punishment" that has a perfect setting that people are too weak to choose. There's a bunch of tradeoffs. Given that, as you yourself keep noting, NYC has a high clearance rate and low crime, but also people sometimes object to specific things the NYPD has done, I don't think it's surprising that people feel like maybe we shouldn't give them a tool that will probably cause SOMETHING to go up when we can't totally predict what that thing will be.
High clearance rate *relative* to other U.S. cities.
My test is very simple: can a resident of a city bike to their local corner store / bodega and go inside while leaving their bike unlocked. If the answer is "no" that means we are below the optimal level of punishment.
I dunno. The reason your bike won't be stolen in Japan isn't because the police are very good at catching bike criminals. It's because the people have lower propensity to steal bikes.
You could argue that crime clearance rate can decrease the propensity for crime but up to a certain point.
While you have a certain admirable consistency about you here, I feel like you're not really engaging with me on the question of whether there are any tradeoffs, or whether reliability issues are a concern in how we deploy these systems, or even more generally the question over whether this is why people oppose it rather than merely a belief that crime is punished too much or too often.
And I've stated before but I think that the reason most people opposite it because being against surveillance is a secular religion.
If it was so terrible then westerners who moved to Singapore would be doing everything in their power to try to dismantle the system. Instead, they forget about it and enjoy the fact that there is zero crime (without ever thinking how).
Vitalik is a great example, he obtained permanent residency in Singapore and lived there for years. But now he is against the U.S. having the same type of safety and order that he thought was valuable enough to get residency for.
It's actually very simple for me: I dislike disorder and crime so much that I am willing to give up a lot to see it go down. Imagine if you were in charge of Venezuela's economy and someone said "there might be tradeoffs" to liberalizing it and undergoing market reforms. The current status quo is so horrifically bad that you would take that risk.
A Tactical Guide for the Highly Intelligent
Marty Nemko
5 min read
A number of my highly intelligent clients can be off-putting. These tactics have been helpful:
Consciously decide when to lift the gas pedal. We all like to be our best, but sometimes, that imposes too big a price. So stay aware of when the risk/reward might be better if you held back, just as a pilot pulls back on the throttle when it’s worth sacrificing some speed for safety, as in a cloud bank.
Yes, it’s cliche, but do listen carefully to your conversation partner(s.) Assess how complex your utterances or writing should be. I often err on the side of simplicity. Even highly intelligent people like to understand without strain. If you’re unsure how to respond, whether in content or level, you might ask a question. Not only does that clarify that issue, it makes the other person feel validated, as long as the question is not asked condescendingly. So, it’s usually fine to ask, “I’m not sure I understand. Could you paraphrase that?” but not, “I don’t get it. Try again.”
My clients and I benefit from, as appropriate, imagining explaining a concept to a bright 6th grader. Often, that’s complex enough. You can always add complexity, but it’s often better to at least start with that as baseline. I have tried to do that here.
Catch them doing something good. Even if they never say something you deem brilliant, if it’s something you think worthy, at least nod and smile if not say something like, “Good point” or “Good job. Thank you.”
Soloing. Of course, when you’re soloing, you needn’t worry about dialing back your brain. So, even if you’re social, spend enough time on solo activities, for example, those that build on your strengths. Possible examples, writing, performing, even gardening.
The latter offers perhaps surprising opportunities for intelligent optimization. For example, I hybridize roses. First, I thought through the best niche — Generalist breeders fail unless they have a large-scale operation. So I put myself in the target customers’ shoes: older people who like the romance of roses but require a small plant because they don’t want or can’t dig deep holes or live in an apartment with only a window box to plant a rose, and they won’t spray with chemicals. Then, I carefully selected varieties to cross. I’ve never taken even a course in plant genetics, but a Google search and $20 membership in the Rose Hybridizers Association, taught me how to choose parents, even which to choose. Then, I created a highly efficient system — including a way to minimize the time-consuming but critical recordkeeping system. I have two roses on the market and nine under test with top U.S. rose distributors.
Kindreds. You’ll be more effective and likely happier when you’re with other high-cognition people. Today, that’s a more acceptable term than “intelligent” or “high-IQ.” Because we spend so much time at work, it helps if your coworkers are cognitive peers. Say, you want to work on Google moonshots. Might you know someone who could introduce you to someone there? Like most highly selective organizations, the applicants that get interviews have usually been referred by trusted colleagues. If not, trying sending a brief, compelling email to decisionmakers, even if a job isn’t posted. The subject line should make clear you’re more likely to be signal than noise. For example, if I were looking for the aforementioned job, my subject line or first line would be “Berkeley Ph.D. in evaluation of innovation with testable ideas for increasing LLMs’ user agency while better optimizing safety vs utility.” Most unsolicited emails get screened based on the subject line.
Outside of work, consider certain MeetUps and online groups, for example, in STEM, debate, or chess. A word about Mensa, where I’ve been a member. I have been disappointed on two counts: It’s more a social club than a think tank for addressing societal problems. And Mensa seems overrepresented by underachievers. I am pleased to see, however, Mensa making special efforts to recruit promising young people — a shout-out to Lana Lee, who is doing great work on this.
Curate your helping. In or outside of work, your help is often most effective when the cognitive gap isn’t large. Counterexample: The large gap between professors’ and students’ cognitive ability inhibits student learning.
Go wide and deep. Most people face a lose-lose choice. If they dabble, they risk being a jack of all trades, master of none. That can be fun but the lack of depth can impede career. On the other hand, in going deep, they too often reach a point where greater depth is needed for career success, but that’s too challenging. That’s a fatal flaw of science museums and of introductory college courses. Those offer enjoyable basics, which hook people, but when they get to, for example, the higher math required in most STEM careers, they quit, having incurred significant opportunity cost. A benefit of being quite intelligent is that you may derive the benefits of going wide and deep.
Forgive yourself. Even high-powered minds screw up and need breaks. I know someone who, to decompress, sews boxer shorts for her boyfriend while watching cartoons. Another pulls weeds. Another reads books for young adults, like The Giver.
Is there one, maybe two things, on this list you want to try, do more of or less of?
The macro
At the societal level, I’d like to see:
Improved restoration of K-12 programs for gifted kids and honors programs at universities. They’ve largely been abandoned in favor of programs for special-needs students. The “best and brightest” are the most likely to cure cancer, better address energy issues, and yes, reduce wars. Yet today, too much of their potential is going unrealized. Programs for the gifted should not be limited to advanced cognitive work but include human issues, like those addressed here.
Media pivot. In earlier generations, excellence was more often venerated by the media and higher education. Today, the plight of low achievers is emphasized and, in some circles, cognitive excellence is viewed as elitist, even racist. Restoring balance is key to long-term improvement in gross world flourishing.
I read this aloud on YouTube.
Marty Nemko is a career coach specializing in people with high cognitive ability. He is the author of 36 books including Work Stories. He holds a Ph.D. from Berkeley with dual specializations: educational psychology and the evaluation of innovation. You can reach him at mnemko@gmail.com
This is missing an introduction that establishes context. It can't capture my attention because for the first several paragraphs I'm just thinking, what is this even about?
> Even highly intelligent people like to understand without strain.
> You’ll be more effective and likely happier when you’re with other [cognitive peers].
> Mensa seems overrepresented by underachievers.
> Improved restoration of K-12 programs for gifted kids and honors programs at universities.
Yeah, full agreement with this.
This is all great advice, and thanks for posting it.
I'd add one thing you didn't address at all:
There are major returns to doing something physical and pushing at least 20% on the "physical optimization" side of things vs being all-in on the mental and intellectual side. Get into a sport, or running, or swimming, or lifting, or some sort of regular physical activity.
Your physical health affects everything, including the quality of your thought, how long you can sustain rigorous thinking, your overall vigor and energy, and of course your morbidity and all cause mortality.
Being sedentary vs exercising regularly is a worse all cause mortality effect than smoking, and being obese has additional ACM increments on top of that.
If you think of the "elements of a life well lived" I think they'd be something like this:
* Good health
* A sharp mind
* An interesting life, full of curiosity and complexity
* Time spent working on problems that fully engage your strengths and are towards some meaningful goal
* Physical vigor and robustness
* Fine meals, expertly prepared, with the freshest and highest quality ingredients
* Robust social lives characterized by truly reciprocal love and devotion and mutual desires for well being and success
* Thriving kids and grandkids
Having the physical dialed in helps most of those! Being physically engaged and adept improves nearly all of those things.
I think there are many issues on both sides of the political isle. However, to be honest, it seems like one side (conservative) is either kept from all the facts (echo chamber) or is openly ok with bad behaviors, policy, constitutional concerns etc in a “ends justify the means” sort of way. It also feels that the “policy” of the right is hard to keep track of. I heard so much from that side about not starting other wars and and “American first” doctrine, yet now that it seems we are doing the opposite, such voices seem to be silenced? I’m trying to approach this with an open mind as I know I have my bias. Background, I was staunchly conservative until Trump and MAGA policies began to shift GOP conservative. Now non registered but leans left and often votes left (but I take it case by case, person by person when I vote). What am I missing. I’d love to hear from conservative voices here in a spirit of understanding.
An awful lot of Americans really, really hate the Iranian regime. Forty-plus years of "Death to America!" rallies have left an impression. And if there are some leftists in the space between Canada and Mexico who aren't terribly fond of "America", those are the ones for whom Iran beating women to death for showing their face in public will be the dealbreaker.
And I'm pretty sure the vast majority of Americans understand that Donald Trump is not going to be impeached for starting this war any time soon.
So, as long as it can be imagined that this will be a Short Victorious War(tm), the pushback will be muted. If you ask people about it, many of them will (and do) note that it was probably stupid and illegal for Trump to start this war without Congressional approval, but they'll also say that they hate the Iranian regime and want to see it fall. And they won't go out of their way to raise a fuss, because what good would that do?
They all get to watch the hated Iranian regime take a hammering, and they can believe that they'll get to see the regime fall, and the ones on the left may imagine they'll also see Trump impeached later on or the GOP take a drubbing in the midterms or whatever. None of that requires anything of them now, so the mostly aren't doing anything now.
When it becomes clear that this isn't going to be a Short Victorious War, that either the Iranian regime will remain in power or that we're being sucked into a Vietnam/Iraq-style quagmire, that may change. But for now, Americans are going to be pretty chill.
You dont think hes impeached by the house if dems take it?
Im not sure how this plays out. I think its objectively bad that he has now had two long planned attacks evidenced by months of military build up, with out even bothering to try to get an authorization for use of force. GOP controls congress so he would have gotten it, and it would have if anything strengthened his negotiation hand in each case.
Its true that the average american would basically be fine with it if both cases turn out fine. But if for other reason his approval is below 30 in 2027, its clearly a good candidate for the legal simplicity of the case.
As for his own reasons, I wonder. It could be cmon no one bothers with these, Obama didnt. Im just going to do what im going to do. It could be the more cunning let them get used to me just blatantly violating the constitution so i can do so again in ways that arent just about process formalities.
He may well be impeached by the House next year. That does not require action or even comment by any protester or dissident *now*, so they're not going to say or do anything now that might harsh the vibe of all the people around them who are just enjoying the show and dreaming of the fall of the Iranian regime.
There is no single right wing block. The older republicans that have supported Bush's Iraq war also support this one, while many on the new right hate this war but don't want to turn on Trump.
The strikes against Iran are not what people in America generally use "war" to mean, and I think they're consistent with how the "no new wars" campaign slogan was understood by his audience, whose central examples are WWII, Korea, Vietnam, Iraq, and Afghanistan. No _invasions_ is perhaps a better translation. Drone and missile strikes are fine, as long as you can credibly claim not to be motivated by "democracy," "human rights," "stopping genocide" or whatever, which are the kinds of things liable to get you to put troops on the ground and start an occupation to ensure those goals are achieved.
I think this is what Hegseth tried to convey with "This is not a so-called regime change war, but the regime sure did change, and the world is better off for it."
How is this different from the NATO strikes on Libya, which Trump called a "total catastrophe?" Iran has 12.5 times the population of Libya and is closer to Europe. If it turns into civil war there's a real risk of another refugee crisis.
> If it turns into civil war there's a real risk of another refugee crisis.
Why? How hard is it to just not take in refugees?
They show up regardless. That means you have guards at the border with guns shooting at them, or you set up camps to house them.
There's an ocean between us. Turning away boats or sinking them is much easier than protecting a land border. Boats are much more visible than people, especially ones that contain a lot of people.
What do you mean by “us?” The US? Im not saying they will come here. I’m talking about Turkey and Europe like what happened with Syria.
Iran is not closer to Europe than Libya. Libya is in fact quite a short sail from Italy, which is one of the reasons why it became such a people smuggling hub.
Well, Iran has a land border with Armenia, which is usually considered a part of Europe, despite being located to the east of the Asian part of Turkey, so that would make the distance from Iran to Europe 0 km, while the distance from Libya to Greece and to Italy appears to be somewhere between 200 km and 300 km.
But I doubt that worry about Iranian refugees pouring into Europe is focused on Armenia.
Here is the video of him talking about it on Face the Nation: https://www.youtube.com/watch?v=4xpqD_SXy34
He says he's in favor of a "surgical shot" to take out Gaddafi and his group, but he was never for "strong intervention."
And that's exactly what happened. The entire NATO Libya operation was seven months and no boots were put on the ground. The US involvement was just 12 days of bombing.
Except the regime hasn't changed, just the guy at the top. Now his son is in charge of the exact same regime.
The Iranians seem savvy enough to read between the lines. The Trump admin's policy is, "Short bombing campaigns project strength and play well to the base, let's do lots of those. Extended campaigns lose support over time and we never win them, let's avoid those."
Last June, the Iranians agreed to the short strike option as a way to get the US off their back for a while. That didn't work, so they changed tactics. Now they have lots of options to escalate this war regionally, and they seem unafraid to do so, because they know that plays against Trump's stated strategy. Meanwhile, the Israelis also want to draw the US deeper into the conflict, as they always have. They're not about to allow the US to bow out after a few weeks of strikes, especially with how long they've been trying to get the US into this thing in the first place.
Sure, quick strikes seem like a viable political strategy, until the enemy hits back and digs in for a prolonged fight. Time will tell, but it looks like Trump has made a profound miscalculation here, thinking he could pull what he did in Venezuela, or do another Midnight Hammer. At the very least, Americans are dying. That's not supposed to happen in the "commit acts of war, but no new wars" strategy. His best bet is to find a quick offramp here, but he no longer controls whether the other two major players in this conflict will let him keep his policy.
That sounds...incredible. Do you have any evidence for this interpretation? I'm looking at current polls, and no matter what you may call it, Americans don't seem to find these strikes "fine".
https://www.reuters.com/world/us/just-one-four-americans-support-us-strikes-iran-reutersipsos-poll-finds-2026-03-01/
I'm basing my assessment on previous strikes, such as on Soleimani last term or the Houthis last year, which people took in stride. You're right that the polls don't look good, but I contend that they're using the question as a proxy for "do you think this will escalate to an invasion – an 'actual' war - and would you support that war?" They get yippy anytime Trump does anything, but I expect that's a transient thing which will pass in a few weeks when/if it becomes clear it's not going to become "World War III" or whatever the current hysteria is.
(Of course, there's also a more secular decline in his approval from his "Après moi, le déluge" economic policy, which also colors some of this polling, and is bound to give the Democrats a trifecta unless he drastically changes course. His foreign policy, whatever it is, isn't going to meaningfully affect that either way.)
>I contend that they're using the question as a proxy for "do you think this will escalate to an invasion – an 'actual' war - and would you support that war?"
You are correct, strikes can lead to escalation, invasion. Therefore, if Republicans voted for Trump because they want "no more invasions", shouldn't they poll with much _lower_ approval for strikes than Democrats?
That doesn't follow. They could also predict a lower probability that these strikes escalate to invasion.
Pick an argument please. You just contended that the "strikes" question was a proxy for an "invasion" question. Now you're saying it isn't.
But torpedo strikes by submarine surely qualifies as war?
Functionally equivalent to an aerial strike.
I think this has a lot to do with the state of online discourse where we are bombarded by ever more sophisticated bots and then have genuine people who are plugged into think tank talking points.
I almost voted for Trump, and part of the reason would have been the anti war rhetoric. Particularly from Vance and Gabbard. And here we are, we let Israel drag us into this war. Which does nothing for us. WTF. And the weird thing to me is all the MSM seem to be in support of the war. Is there anyone in congress against it? Rand Paul? Why this almost universal support in our congress for Israel?
Congressional Democrats are like "while we fully agree Iran is really really bad and (list of horrible things khomenei has done) Trump is not explaining adequately why he's doing this, what the plan is, and also isn't Congress supposed to be the branch of government that declares war???" Source- an email I got from my congressional representative to the constituent email list. So no media filter there. Sample size of one, but I think mine is considered a fairly standard center of party Democrat.
I don't think Israel dragged America in. Israel could have and would have done this on their own if they decided it was necessary and saw an opportunity, if America hadn't been ready they wouldn't have waited for them. (Proof: last June.). My pet theory is that this is somehow connected to Trump's personal financial interests in a way that will probably be exposed at some point in the future.
OK the Dems can and will say all these bad things about Trump, and he has no plan and didn't ask congress*. And still support the war. IDK.
My take on Israel comes from the breaking points podcast. Where I say a clip of Marco Rubio saying (roughly), "Yeah we knew Israel was going to strike Iran and we knew Iran would respond against the USA, so we went in there preemptively and took out there missiles and such. So they couldn't hurt Americans." That counts as 'dragging us in' in my book. I don't think it's some Trump evil plan, but that's just me.
*As I continue to say the biggest prob I see with the way our government works is that congress has ceded too much of it's power to the Prez. If they wanted to take back that power they could. And control the war and the budget and tariffs and... They are wimps and we need better congress people.
It's difficult to understand these justifications. If Iran was planning to attack the US first (even if Israel attacked them first, we could at least claim we weren't fighting Iran at the time) that would have given the US a casus belli for war. It's an absurd response to the question, "What's your justification for this war?"
"They were going to give us one, but we beat them to it and gave them an excuse to retaliate whole depriving ourselves of a reason to fight." This is a far cry from Lincoln, setting up the South to attack Sumpter - which they obliged him - and give him a reason to mobilize the public against the aggressors.
You think the mainstream media is entirely in favor of the war? That is not my impression.
OK Good, I don't watch MSM, but just listen to a few podcasts. So who's against it?
The NYT headlines on the war have been mostly negative, focusing on the spreading side effects of the war and the fact that Trump doesn't seem to have a clear justification or goal for it.
Sure negative Trump coverage is expected, they may still support the war. IDK.
They were never going to be for it. This is the NYT.
Trump could physically charge into a burning building, walk out with an armful of kittens and babies he saved from most certain death, and the NYT would spin it as a negative.
Then why the heck did you think the MSM would be supportive of the war?
You seem to have lost track of the actual question being discussed here.
This is about my feeling as well (bias warning: started as a cross pressure moderate, drifted left over ~10yrs to be a classical left-liberal).
In general both parties seem at their worst when bowing to their ideologues (Trump for the right, progressive tastemakers for the left). But the centralization of power on the right seems to have legitimately excluded the conservatives that I could imagine making common ground with, while I think much of the mainstream Democratic Party still has a lot to offer people of all political persuasion.
I will agree with Shaked K. in this thread that Blue State (city?) governance produces a real image issue for the left at the national scale.*
*A little out of scope for this comment but I think that the realities of national politics tends to curb the worse impulses of the left in a way that’s not symmetric to the right.
About your asterisk, I think the bigger curb isn't just the realities of national politics, it's that it's easier to get competent high level administrators for big programs run by Washington than for state or local issues.
(That said, mostly I just don't see what the Democratic party has to offer me anymore. I used to like industrial policy or infrastructure projects or Medicare, but the major programs for that all seem so consumed by waste and fraud I've changed my mind and would prefer tax cuts now).
I've heard from people on twitter (el gato malo) that there's open war in the GOP netizens, in terms of isolationists versus interventionists versus Trumpian-uber-alles supporters.
This...does not reflect my understanding of Republican politics.
To start, there are pretty deep fractures in the Republican base over this. You can see Tucker Carlson condemning the Iran war here (1) and you can see Ben Shapiro taking some pretty pointed shots at Carlson here (2). I think a lot of political actors are beginning to prepare for the post-Trump era. He's old, obviously slowing down, and the post-Trump presidential campaigns should begin after the midterms. It's very unclear where the party will go though; there's no unanimity among the pundits, politicians, or candidates.
If the Republican party seems contradictory on this, it's because internal factions within the party are fighting over narrative control.
(1) https://www.youtube.com/watch?v=JZxj_c0g5lU
(2) https://www.youtube.com/watch?v=LDBM_SepNWI
I'm in the exact opposite boat from you (lifelong Democrat until 2025) and I have the opposite impression (for example, most people I've talked to on the left seem sure that Milei destroyed the Argentinian economy, and every supposed moderate democrat immediately jumps to the left five minutes after being elected).
It's easy to point at bad things Trump does (I probably agree with you on most of them). But I still lean right overall, because I think the left would be significantly worse in power. We don't see it partly because they're not currently in power (and extrapolating from Obama or even Biden to current Democrats would be like extrapolating from Romney to Republicans), and partly because Democrats' mistakes are more invisible - going too far in letting process or bloat accumulate is generally a bigger issue than acting too severely, but it doesn't cause sudden problems so it mostly gets overlooked.
We also just have an empirical comparison between red and blue states - I think it's pretty hard to argue that California or New York are well governed compared to Texas and Florida (you get similar levels of service at best with significantly higher taxes).
None of this is to say there aren't real problems with the current version of the Republican party. But I think it's more salvageable than Democrats, so I prefer trying that instead of going over.
>going too far in letting process or bloat accumulate is generally a bigger issue than acting too severely
No, it's generally less of a problem. That why there are red lines in the constitution. Torture *is* worse than dust specks.
The current administration attempted to enact the largest tax hike of my lifetime, and the only complaint coming from the party about it is that the Supreme Court stopped him from unilaterally creating a massive tax hike by Presidential diktat. Come on man.
I agree that the tarriffs are very bad in multiple ways, and are a prime example of the ways in which Trump is often pointlessly destructive. That said, Republicans are still broadly speaking the lower-tax party (especially at the state level). More importantly, they're still broadly deregulatory (Trump has a few big pro-regulation areas about which he is very bad, but he also put the 2/10 for 1 cancelled regulations per new regulation rule, which makes up for a whole lot of bad stuff).
Going forward, I think the best political strategy is to try to empower the traditional/Rubio wing of the party over the populist/maga pro-tarriff wing. It's not that I disagree about the destructive decisions Trump makes (again, I opposed him in all three of his elections!), I just think it's unfair to compare him to a neutral placeholder instead of the likely alternative.
> That said, Republicans are still broadly speaking the lower-tax party (especially at the state level).
Again, quite literally, the largest tax increase of my lifetime was enacted by Donald Trump, and approximately 0% of the Republican Party or its base pushed back on him about it. You cannot say you are anti-tax at that point, I'm sorry, I don't believe you. You had the most obviously bad tax policy imaginable, enacted without appropriate mechanisms (e.g. Congress writing a bill), and they just bent over and took it. It's fucking tax policy! It's not some weird deregulation of pharmaceutical laws or something, "don't take my money" is the most vulgar form of conservatism there is! It's the conservative equivalent of lefties complaining about billionaires!
It's over, MAGA won, you can't cite 40 years of effective libertarian-leaning policy when the current leader of the party has a meltdown about a Canadian ad regarding Reagan opposing tariffs and everybody in the party just goes along with him!
> I just think it's unfair to compare him to a neutral placeholder instead of the likely alternative.
The "likely alternative" is Vance, not Rubio, who is also, err, not pro-deregulation.
e: Sorry, wait, if you meant likely alternative as in Biden or Harris or Newsom... bro! Not a single Dem in my entire life has raised taxes as much as this motherfucker, come on!
Are you talking about the tariffs? Otherwise I'm confused.
Yes, tariffs are a form of tax. Is that a controversial statement?
The current administration is an incompetent mess. Just like every other Trump venture. He’s just breaking shit. What has he done that will have a positive long term impact? I can point to him defunding and destroying immensely valuable institutions endlessly.
There's some obvious things, like the ongoing Iran campaign or various deregulation measures, but those are controversial so let's put them aside for here. I'd argue that (a) some amount of creative destruction of institutions is necessary (to quote Elon Musk, if you don't regret at least 10% of the things you removed you're not removing enough things), but mostly just (b), I think Trump is bad but less bad than a typical Democrat would be. If we could go back to the Bill Clinton Democrats - something you can draw as a normal baseline - I'd take that over Trump, who really does as you say do a lot of stupid or destructive things. But the current energy in the democratic party is on far more destructive left-wing policies than that (you can look at how the places where Democrats are in power to get a sense of the results), so I now think Trump is less bad than they would be.
(Note that this is new for me - I didn't think this way in the 2020 or even 2024 elections, and I agree with most criticisms of Trump. But seeing the Democrats' reaction over the last year has pushed me over the edge).
> some amount of creative destruction of institutions is necessary
It's only _creative_ destruction if you actually _create_ something, otherwise it's just _destruction_. Trump has, in just one year, done more damage to America's soft power and reputation as a reliable economic and military partner than any other president managed to do since the end of WWII, by a wide margin.
> to quote Elon Musk, if you don't regret at least 10% of the things you removed you're not removing enough things
Some actions can't simply be reverted in the next software release. Once you've openly and explicitly threatened to invade an allied country for non-sensical, petty reasons, there's no going back. Sending Rubio with a superficially soothing speech to Munich won't undo the damage you caused.
> It's only _creative_ destruction if you actually _create_ something, otherwise it's just _destruction_.
The existing systems need to be thoroughly destroyed before new ones can take their place. There will inevitably be some growing pains.
First off, this is wrong. Creative destruction happens through better systems displacing and replacing existing ones, not by deleting the existing systems and then figuring out how to fill the resulting vacuum.
Second, Trump isn't exactly the kind of guy who has the experience, the will, the attention span, or the consistency for the long, arduous, and unglamorous process of building new, competent institutions, nor does anyone in his administration [1]. It just doesn't make for "good television", and burning bridges or humiliating partners is much less boring.
[1] Maaaaaybe Rubio does. But as long as he has to bend to the daily whims of his boss, he won't get the chance to do so.
That's cool you can think in the most banal abstract terms that don't apply to what is happening in reality.
Speak in concrete terms and point to verifiable things because you really said nothing here.
What about the outright grift and corruption? The tariffs which were paid for by USA consumers? You think we should have eliminated waste in the government. For the deficit? Then why did we give tax breaks to the billionaires? You are aware their effective tax rate is lower than any other citizens?
Did you catch the Democratic elected officials trying to invoke "secret hidden signals" to countermand Trump's position as Commander in Chief? One of them got his rank/pension busted down by the DoD. That's far from invisible.
You’re jumping the shark on this one. Probably time to dial it back a notch.
I did not, what is this? Not finding it online
https://www.foxnews.com/politics/arizona-sen-kelly-sues-hegseth-over-military-pension-cuts-following-video-message?msockid=14ac9e0ef2c66ca4260c8901f3726d46
Watch the video (not sure if it's linked here, I just pulled Kelly's latest salvo -- I can find a link if needed).
Okay yeah not seeing secret hidden signals here? This fight was pretty blatant on all sides.
https://boriquagato.substack.com/p/coloring-outside-the-lines-of-color
(skim down to where Senator Elissa Slotkin starts talking. It's in her words.) You don't need to look at el gato malo's philosophy to look at their own words. "Don't give up the ship" is weirdly put, and strangely put. The whole thing reeks of coded messages (have you seen the video where Nancy Pelosi's brainwashing gets triggered? It's hilarious -- she literally reboots).
Pretty sure this is just an extremely credulous reading of the administration's attempts to prosecute Senator Mark Kelly: https://www.bbc.com/news/articles/cvgk4vl30vko
What? How "credulous" do I need to be to see them court-jumping from "let's prosecute this in civvie courts" to "let's use the UCMJ and the JAG lawyers" to "ya know, let's just bust him down a rank and call him slapped down"?
I was mostly interpreting:
> Democratic elected officials trying to invoke "secret hidden signals" to countermand Trump's position as Commander in Chief
Which is either a credulous reading or a sarcastic credulous reading, I couldn't really tell. My apologies in the latter case.
This is the poster who claimed we haven't had free elections in parts of the country since the 70s and then provided 0 evidence when asked.
Many popular conservative pundits and influencers will say whatever they’re paid to say or will go along with the party no matter what. They aren’t principled and just serve a function of herding the masses. Trump has betrayed his base on virtually every issue and some supporters are slowly starting to realize that.
That said, it’s not like the left is somehow more principled. They never cared about the lawfare against Trump, the denial of due process for J6 offenders, illegal censorship during the 2020 election, and countless other issues. They are no more well informed - take the ridiculous claims made during the Kyle Rittenhouse case for instance. Or basically anything related to woke ideology.
What are the riduculous claims made about the Rittenhouse case?
Without relitigating the entire thing, there were all sorts of nonsense takes going around about it somehow mattering that his home town was on the other side of a state border, or that he was a murderer or a terrorist of some kind. In reality there was abundant video evidence of it being textbook self defense.
Given that he placed in himself in danger by choosing to attend the riot.
I don't recall the argument from.terrorism. I do recall the argument that avoiding trouble is the first line of self defense.
> denial of due process for j6 offenders
What's this one? I haven't heard this claim made either way before
J6 folks sat in prison for so long that their sentences were commuted, in general. 6 months? free to go, you already served that before the trial.
Folks are rightly upset if they want to complain about "this took too long" -- this was "bill of rights" level egregious conduct by the Biden Administration.
That's not "denial of due process". The criminal justice system is indeed too slow, and people will take plea bargains because it's quicker, but that LONG predates J6. J6ers getting commuted by Trump means the deviation from normal process was in their favor.
Spot on. “That shouldn’t be what happens” /= Due Process violation.
By prison do you mean pretrial detention?
(In general I think the fact that we take years to do trials instead of days or weeks is a pretty big travesty. I don't know if it's specific to j6 people though? Maybe if the government didn't let any of them plea out for political reasons)
sorry, I should have said jail rather than prison (there is a difference), my bad.
In general, it doesn't take years to do criminal trials (and when it does, it's because someone wants it to take that long, often the defense -- if Chauvin wanted a longer trial, I wouldn't blame him -- police in prison have a rough and short lifespan). These were excessive, to the point that the process was part of the punishment.
Iirc the norm is six months to two years depending on the crime. I can believe the government prefers to stretch it out in cases they want to be mean.
I honestly think both sides have approximately equal percentages of low informed voters and ends justify the means voters, but there is an important difference between the parties. The Republicans have many more politicians that are low informed or ends justify the means
Does this mean you should stop being a Republican? It means the opposite, at least if you’re in a closed party state. You should stay a Republican and vote in primaries for candidates that are well informed and do not believe the ends justify the means. Democrats primaries are not nearly as important as Republican primaries. You can still vote Democrat in the general if need be, but you are an important force in making the Republican Party back on track. At least that’s been my plan for the last decade.
Are you talking the economic terrorist Republicans? Because I think we pretty much got rid of them, and the new economic terrorists are on the Democratic side.
*Yep. "economic terrorist" was bait, read out by an Irish-American Representative. When folks tried to call him on it, he mentioned that he was just quoting the former Republican Secretary of the Treasury (a constituent).
Steve Yegge just announced Wasteland, which is his federated version of Gas Town, which in turn is the most aggressive multi-agent coding tool I've ever seen. It doesn't stop at writing code, it will merge PRs to main, by default.
https://steve-yegge.medium.com/welcome-to-the-wasteland-a-thousand-gas-towns-a5eb9bc8dc1f
This is exciting for me personally since the backend is now running on Dolt, so we're seeing tons of new usage for the database and DoltHub. But in terms of whether the idea of federated multi-agent AI rigs working together is insane, or useful, or doesn't go nearly far enough, I am very uncertain. I'm curious what ACX readers make of it, especially if you've tried Gas Town yourself.
I regularly see leading models making stupid mistakes in code, so yeah, no thanks.
They're still great timesavers, mind you.. just need a bit of supervision.
I'm not nearly rich enough to try GasTown. I feel like you have to be some kind of eccentric, successfully-exited founder to be willing to flush the amount of tokens really running it requires. For my scale of projects, an interactive Claude Code session is plenty productive. Frankly I think GasTown is going to quickly become an obsolete curiousity of an intermediate age, and all functionality it provides will ultimately be subsumed by the models and model providers themselves. You can already see this coming with Claude Code swarm mode, Codex multi-agent management, etc.
edit: I see swarm mode is officially "agent teams" now that it's out of stealth
I barely even use Claude Code and I still constantly hit the usage cap ($20/mo sub). I really wonder how all the hypesters and vibecoders are doing it. Are they really all paying 200+ a month?!
I've read all the Gas Town blogs and now the Wasteland one (thanks for the link, BTW!). I have not used Gas Town and am not likely to, although I am a heavy Claude Code user.
Gas Town has some interesting concepts (its reliance on Beads for logging issues discovered during implementation for one, although Beads wasn't working well for me on Windows when I tried it a month or so back) but to me it's a very Ptolemaic system -- epicycle upon epicycle to try to keep the underlying models coherent and on task indefinitely.
Does it work? Maybe. I will be more convinced when I see significant software produced using it that is not itself or its own tools. For now, my personal sweet spot is setting up automation workflows with Claude Code that involve up front detailed specs with human review gate checks. This approach uses my human judgment and knowledge to keep the workflow moving in the desired direction rather than a crosstalking collection of agents roleplaying a corporate development structure.
I don't know if my way is more effective, but it's simpler and more scalable for what I'm using it for.
I know nothing about any of this, I just can't figure out what level of irony you have to be on to name something "Dolt".
It's an homage to how Linus named git but the reference does go over most people's heads.
https://docs.dolthub.com/other/faq#why-is-it-called-dolt-are-you-calling-me-dumb
My wife would certainly add "Dolt" to her list of top tone-deaf product names, which currently includes Claude -> "clod" and "Soylent" -> "soil-ent".
Don't forget Discord (an ugly sound, dissension and strife), Palantir (a corrupted tool of the great Enemy), and nVidia (invidious - offensive and unfair, tending to bring resentment)
The world will end, not with a bang, but with a
git push origin master
you forgot the -f
This is why you make master/main a protected branch!
Sounds like *someone* isn't 10xing their productivity with [insert AI for which your employer bought an enterprise license] here!
I read through the Gas Town post but have not used it. I don't think I can justify spending money on it out of morbid curiosity.
When I read about these sorts of things, I often think the same thing, which is that I have a general theory of what my job is, and apparently it's very different than Yegge's theory of the same.
Most reasonably-complex software systems are much too large to fit into your brain at once, unless you etch them into your long-term memory. Most of my job is to deeply understand systems so I can know what is reasonable to do with them.
If I no longer have to read through the code a bunch to make changes, I feel like that's getting efficiency today at the cost of efficiency tomorrow (or forever) by taking away my understanding of the system.
I'm sure some people will say "You don't need to understand the system - the LLM can read the whole thing and understand it in seconds", but that doesn't really stop me from getting the LLM to make terrible mistakes because I ask it to (and cannot recognize them as mistakes during review).
Also, I write a disproportionate amount of the incident reports for my team (mostly out of, again, morbid curiosity) and a lot of things that cause incidents are quite subtle or even arguably-sort-of-not-wrong.
I still think these are useful tools. I use them regularly. But they give me wrong answers a lot, even with the latest models. And they struggle when I give them more complex tasks (my current attempt is along the lines of "find me any code that isn't technically dead but seems like it really shouldn't get hit ever", which seems to be too much because in most cases you need domain knowledge to appreciate what "shouldn't" happen).
I've often felt the same way.
Thank you for this. I have been reading Yegge on and off since "The Kingdom of Nouns" and have a strong bias to believe he knows what he is talking about more than I do but this is a well reasoned counterpoint that matches my own experience.
I'm still confused about what is actually being built. As far as I can figure out from the Wanted board, Wasteland is mostly just building Wasteland so that it can build Wasteland?
Yeah, the tendencies towards LARP-ing, gamified badges, and "recursive-self-improvement" have crowded out anything else this system might do.
Also: he says 1500 pull requests is "a lot more than most companies have done in the last 2 months" - but I managed 350 in February with a normal usage of Claude Code / ChatGPT Codex.
I have not tried Gas Town, excited to hear about Wasteland and always interested in what Yegge is doing.
I finally got around to using Claude Code over the last few days and...I'm pretty ok with letting this thing run loose. I think it will make mistakes, I think there are types of tasks I wouldn't turn over to it but...it's not significantly dumber than me or my coworkers. It was like...I basically ran it in another monitor while I was in meetings all day, checked each code change as it was made, didn't see anything wildly stupid, and approved it. That feels "safe" and I'll probably do it for the next 6 months but...man, I feel really unnecessary in this loop.
I can totally buy a fleet of coding agents doing crazy work.
ACX readers, what is your current AI setup, accounting for a wide range of use cases and risks? So many recommendations, so much confusion. Sample idea from my colleague:
Level 1, top privacy: locally run models for highly sensitive stuff (e.g., Llama 8B + LLM Studio)
Level 2, higher privacy: OpenRouter + pay-per-use models, email/payment method/prompts without personally identifying information. Alternatively, Perplexity subscription for convenience.
Level 3, low privacy: Claude Opus and Gemini coming with Google Workspace for public-facing work, email/payment method/prompts with identifying information.
(Plus, if you’ve overshared in the past - non-obvious data poisoning)
None whatsoever, though I'm considering an Anthropic subscription if Dario holds firm.
I can't use any of the public models for my work, because everything is CUI or proprietary. The few times I've tried to use the company's approved in-house model, the results have been worse than useless except occasionally for entertainment value.
And I haven't found any really compelling use cases in my personal life, at least from what I've seen from colleagues who have tried it. An amusing toy, but too unreliable to be trusted with anything important and I don't need it for the unimportant stuff.
Also, I have a philosophical objection to sending money to Sam Altman, and I'd have to think real hard about Elon Musk in this context.
I have a $20/month Claude subscription, a $20/month Google AI Pro subscription, and a ChatGPT Pro subscription I got for free. I tend to use them like so:
- Gemini for things that involve visual reasoning eg, validating structural design questions, complex multistep identication-and-reasoning from images, etc. Nano banana (2) is also very good at image generation, on the rare occasions I want that.
A typical example (ID this connector - note that Gemini got it wrong at first blush, but was helpful later): https://gemini.google.com/share/e5e73877866a
- Claude for coding related queries, sometimes in the web UI and sometimes Claude Code, both via the terminal version and the vscode extension. And the occasional fun query when I want "someone" to bounce a strange idea off of.
Code related (help write a selection shortcut handler for Autodesk Fusion): https://claude.ai/share/2a488c49-e83b-4658-a262-990bc857095e
Fun: https://claude.ai/share/24f2d23e-0a00-4d9f-a999-f9b9c1c2201c
- ChatGPT for the occasional quick query, ie the name for a specific part, help recalling cases I can't remember the specifics of, etc, plus the occasional Deep Research report. I found the 5.1 personality so annoying it soured me on the whole brand, even after I updated the default. Apparently 5.3 is supposed to fix this, but Claude is just nicer to interact with, overall.
Basic example (What's that thing called?): https://chatgpt.com/share/69a9e9fb-5e30-8008-a28f-df27de71c05c
Deep research example (I wanted to validate the recent surprising claim about feeder mice being one of the most farmed animals in the US): https://chatgpt.com/share/69a9ea35-4ab8-8008-b492-42cc8a57fa26
Almost daily use for tech work, via API pay as you go, what you'd probably call level 2. "Review this code", "how would you implement <whatever>", "explain this error message", "translate this sentence", etc. It has mostly replaced StackOverflow and online manuals for me. Just getting into more automated tools like Amp code and Claude Code. I also use the API to ask general curiosity questions about anything from Sanskrit etymology to elderly relative's medications.
On the phone, random questions go to the free tier of the chatgpt or gemini apps (level 3 in your list).
OTOH, I regularly turn off "AI help" features in apps, browser and OS's. I want AI to answer when I ask, not to do shit on its own.
Level 0, nothing at all, and telling Copilot to go to hell every time it tries to pop up and be helpful. Begone, bastard offspring of Clippy!
I am forever going into my settings on different devices to shut AI stuff off.
At work, I use the Copilot plugin for VS Code. My employer's IT and legal departments have signed off on us using that, and I figure it's their call whether it's secure enough for the company's purposes. I make relatively light use of it, but do find it somewhat useful for intellisense-like purposes.
Outside of work, I mostly avoid it.
Level 1 and 3: I have 128GB of RAM and can run Step-3.5-Flash, which appears to be a distillation of OpenAI o1. Level 2 seems completely pointless.
I was about to get a new Mac for Level 1, but my Apple Store has cleverly stopped selling Macs temporarily until the new ones come out in a week or so. The man is always trying to get me down.
I got a mini a month or so ago. Have they stopped selling even those?
I didn't check, I don't have a spare monitor at home so I didn't want to get one.
Just wired up a level one machine
I don't use it.
I do unfortunately have to use Windows 11 sometimes, but I ripped out all the Copilot crap I could.
Do ya'll think Iran executed all the protestors from last month they had sitting in prison? On the one hand, I haven't heard anything to indicate this has happened... on the other, it seems like a really obvious move during an attempted coup (and what they might have been planning anyways).
The legal opinions around the OpenAI contract are interesting to read, but there's one very fundamental issue that no amount of lawyering can reassure, right?
The crux of the original conflict was that the DoW insisted upon the right to cross at least one of Anthropic's two red lines.
No matter what assurances anyone from OpenAI or DoW gives, it is unquestionably the DoW's understanding that the OpenAI contract language is more permissive than their contract with anthropic would have been. Otherwise why they bother with this whole rigamarole? Am I... Somehow mistaken on this? I feel like this is the point that needs to be pressed within OpenAI.
If Sam Altman believes that the red lines as would be understood and interpreted by a normal human off the street are in place, while the DoW is nonetheless satisfied, then he is asserting that he is somehow hoodwinking the DoW in plain sight, no? Doesn't pass my smell test.
IMO Sam Altman knows that the red lines have been crossed, the DoW knows that the red lines have been crossed, OpenAI is okay with losing a few customers over this, and OpenAI is just making empty statements and doing security theater to try not to lose their employees. Dario Amodei is a biased source but I really agreed with his memo
https://pbs.twimg.com/media/HCmdjFGXwAAPI3d?format=jpg&name=4096x4096
I think Sam Altman is talking out of both sides of his mouth here, he pledged to stick to the same red lines as Anthropic AND grabbed a bunch of Pentagon business, all within 24 hours. I think he's just saying whatever sounds good, like many other CEOs.
It's most likely that the new agreement was due to some perceptual or contractual loosening, but it is possible that the DoW straight up chooses to treat Anthropic and OpenAI differently because OpenAI has more political favor with the admin and the reverse for anthropic.
I think you're probably right, although I also wouldn't put it past the current administration to be so stupid and narcissistic that they would switch just so it looked like they "won" and/or be tricked by OpenAI.
Either OpenAI's contract let's the DoW do more stuff than Anthropic's or it doesn't.
If it does, well, that explains everything.
If it doesn't, DoW might still have gone for it due to all the donations OpenAI people have given to Trump for example, and more generally the ways in which OpenAI has sucked up to them.
Both possibilities seem plausible to me.
Another important piece of context is that Sam can just say one thing to the public and a different thing to the DoW behind closed doors, e.g. "Don't worry all this talk of red lines is just for show, we won't complain if you quietly change the terms of this contract later and/or blatantly violate it"
I agree with both this comment and Yug Gnirob's comment below.
It's either that the contract is more permissive, or there's some other explanation for why this all happened (likely from the set of options presented in the replies.). In either case, public reassurances about the quality of the safety stack or legal language are noise when it comes to resolving this core concern.
I myself don't have a strong feel for which one of the offered up answers is the real one, so I'll refrain from making any guesses here. Just that the official story being presented in public does not add up no matter how you cut it.
You're assuming the DoW is following their stated logic, instead of looking for an excuse to preserve their ego. They're the kind of people who would take a worse deal so they could lie and say they made a better one.
Yeah, I mean, look at how tariffs have been handled to see how much the administration values consistency.
Tariffs are a point in which "consistency" is very clear. It's just that the agenda is very different than you think. Chaos is a means to an end, a very particular end.
What is this very particular end?
Deflating the stock market on Trump's terms. It's in a massive AI induced bubble, and if Trump deflates it, Trump can reflate it as he pleases (a managed recovery if you will).
It's a high stakes ploy (and there's other agendas at work, too), but Trump kinda likes those.
What's wrong with using interest rates?
How long do you think the Iran war will last?
It depends on whether Trump puts boots on the ground. IRG says it intends to fight to the last man. I think the Pentagon wouldn't want to invade. Trump made a spontaneous Truth Social post the other denying that we're running out of missiles. Funny thing is, no one had brought this up as a possibility. So I have to wonder if we aren't running out of missiles?
I don't think there will be a ground war in Iran. If we run out of missiles, and Iran runs out of missiles, it will probably fall into a stalemate. Israel may keep it going, though.
On the principle that it’s always easier to start trouble then stop it, longer than most people think. Much longer.
I'm very new to this, but a few random factors that could prolong things on the Iranian side:
1. US is supporting the Kurds, who also get support from Iraqi Kurdistan - there aren't enough militants to actually threaten a major takeover, but having an active Kurdish insurgency in the Northwest could keep some kind of conflict ongoing for a long time.
2. Random other insurgencies like the People's Fighters Front/Balochistan stuff.
3. Khamenei succession might end up being chaotic - probably some messy internal politics that it's difficult even for an expert to understand, let alone a noob armchair observer. Khamenei's son seems like the likely successor, but the irony of a revolutionary republic founded on overthrowing a hereditary monarchy doing father-to-son succession will surely raise some thick Persian eyebrows!
4. The IRGC-adjacent Basij militia seem to be a factor that stops the "fairytale" idea of a liberal uprising - a big chunk of them are pro-regime thugs who are embedded in society... but I've got no idea how deep their loyalties run
5. The Artesh (conventional army) is likely to gain power relative to the IRGC, as the strikes continue to decimate IRGC infrastructure and leadership. This might mess up domestic power structures a bit, and give more power to the (less extreme) Artesh. Can't see this leading to a coup or anything, but having two non-aligned militaries could make Iran a lot less stable.
This is all true, but there's an alternative story, where Shia Islam makes a play in the various countries allied to the US, or nearby, and the conflict spreads out of control. We don't know what the next Supreme Leader of Iran will do, but with the protests around the region this week, there seems to be a lot of potential unrest that could go against the US presence. Lebanon has joined the fight. The Houthis have vowed to shut down oil exports in their region if the Saudis join the war.
The US strategy seems to be that if they create hardship, in an unpopular regime, ruled by autocratic dictators, who are a minority presiding over a silent majority that has a long-standing hatred of the government - that this combination of factors will be enough to destabilize the regime and lead to a popular uprising.
Okay, but that describes every US ally in the region, only moreso. And now none of those regimes are getting oil income.
Proxy groups like Hezbollah and the Houthis survive because Iran funds and supplies them. The Houthis are probably not building anti-shipping drones themselves. It's hard to know how effective these groups will be if Iranian support becomes unreliable.
Seems clear from what we've seen over the past year that it's a more nuanced relationship than a strict proxy fight. Iran pulling back would certainly hinder the Houthis, though I don't think it's fair to say Iran totally calls the shots of what they do or don't do.
Meanwhile, China seems interested in using this conflict to their benefit. As does Russia. I'm not keen on discovering what creative surprises they're currently working on to make life more difficult for the US in that region.
Many countries, China, India Japan South Korea have limited stock of oil and rely on continuous import from the Gulf. They are going to start yelping pretty soon.
Probably a few weeks for the main bombing campaign, with some long term low level warfare (e.g. things like arming the kurds). The navy, main missile factories, and much of the governing infrastructure has been taken out ahead of schedule Afaict, without a land invasion that's probably close to a full air war victory.
I doubt it will go on for long. When it's a liability domestically, the Trump administration will change it's tune. American casualties will accelerate the process. It's a terrible time, politically speaking, to be getting American soldiers killed in a war of choice. And this was a choice: Iran was no threat to civilians residing in North America.
So far American casualties are very low (less than Pakistan's!), because we've been avoiding "boots on the ground", and Iran's capacity to strike outside its borders has been so diminished.
The US wanting out won't be enough to get out, unless it's also willing to withdraw significantly from the region
Iran correctly sees the US bases in the region as a significant threat and is probably unwilling to settle for "okay we'll stop so we can do it again in a few years"
I can't see the US pulling out if Israel is still fighting. Bibi has wanted the US in this war for 40 years (his words). He's not about to let us walk away after a couple weeks of bombing. They dragged us into this, they're not letting us back out. There will be escalation, it's just a question of what that looks like.
Iran's government is a threat to civilians residing in Iran, where tens of thousands were recently murdered. It doesn't look good complaining over a few American deaths.
I'm highly skeptical of this claim. What verification do we have that it's legitimate? It seems to have come from unnamed Israeli sources, but I'm open to better sourcing if you have it.
Do you realize how logistically difficult it is to kill that many people in so short a time? And how difficult it would then be to verify that number that seems to have come from nowhere? This looks exactly like the kind of talking point that has been used for decades to get people enthused for every other war in the ME. Can we please not fall for it again this time?
What were know from that period is:
1. There were lots of protests (both pro/anti regime)
2. There were professional agitators among the protestors
3. Those agitators were armed with fully automatic weapons and targeted police and firefighters
4. The Israelis admitted they were involved
What this looks like is that there was an attempted color revolution that the regime violently suppressed. It looks like lots of innocent people probably got caught between these two sides fighting for power. Legitimate numbers I've seen are in the low to mid thousands. That doesn't sound like a regime being careful with the lives of its citizens, but like panicked people doing what they can to stay in power.
I'm not here claiming the Iranians have some great track record of human rights. But they didn't wake up one day and decide to start slaughtering people days after Trump publicly declared that one of his red lines for intervention was if the regime started mass killing their own people.
This was never about the humanitarian situation. You know what because the US is allied to the Saudis, among others. The US entering this war will almost certainly cause more Iranian deaths than the fake tens of thousands being used as a pretext for war. Just like in every other conflict where this tactic has been employed.
Official estimates from the Iranian government are over 3,000; various human rights activist groups claim to have verified 5,800–7,000 deaths; other estimates are higher. "Tens of thousands" is sufficiently disputed that I'll stop using that, but "ten thousand" does not seem to be an exaggeration.
10k vs 20k does not change the moral calculus.
The motivations of Trump are, ultimately, irrelevant to the moral question here. I'm pleased just to know that most of the people who made the decision to murder those people are now dead.
I'm not certain about that. Let's say 8k, which is on the high end of the credible sources I've seen. Of those, I've never seen a breakdown of how many were killed by the regime versus the terrorist groups gunning down the police. Is it 50/50, 90/10, 10/90? There's no answer to this other than vibes/bias at the moment. Partly because the regime isn't interested in letting in human rights groups or free speech. That's not cool, but then again the two countries that just openly started an aggressive war against them bragged about sending terrorists into their country to kill policemen. My country has done far more with far less justification .[EDIT: here I'm referring to the speech restrictions, not the killing of protesters; the US is far behind in that category, with a much smaller - though nonzero - body count].
Whatever the balance, I doubt this intervention will cost fewer lives on balance than the status quo. I don't expect the nation that sent terrorists to kill firefighters in Iran to be careful about civilian casualties. I've seen the videos of mass bombing over Tehran. Reportedly they're planning to send gravity bombs next. Not exactly targeted strikes.
Meanwhile, my government lies to me. All the time. Never more consequentially than when they're lying us into war. Vietnam (Gulf of Tonkin), Iraq war 1 (babies in ovens), Iraq war 2 (WMD), Syria (chemical weapons), and on and on. Later we find out that they used our compassion to get their war. Large numbers of people die so the US can ride in on a white horse chasing windmills.
I'm out. Call me an isolationist, but better that than following these warmongers who always find some new excuse, but never an actual cause for war.
America recently moved on from blowing up an elementary school to just outright terror bombing Tehran, while aiding genocide in Gaza
We did carpet-bombing in WW2. I don't think we've done that in Iran, partly due to increasing precision.
Sensible people have always denounced carpet-bombing.
Calling the Gaza campaign genocide is an ugly smear. It should stop. The entire city had to be razed because the entire city is in service to terrorists. The violence is terrible, but appeasement and doing nothing did not work. October 7th proved that. It was time for Gaza to go.
I'm fed up with this ridiculous pretense that Hamas is a threat to Israel. Hamas is a pathetic third world mob with kalashnikovs and fireworks. It should not be a threat. The rational thing to do, even just economically let alone morally, would have been to build a figurative wall around Gaza and otherwise let them be.
October 7 was carried out by a mob with small arms, which would have been stopped by something as cheap and simple as a minefield. Somehow it took Israeli helicopters half a day to take off. Israel ignored an explicit warning by Egypt on October 4 that a big attack was about to happen, as well as many other warning signs.
The lessons Israel should have learned are: Set up a minefield and other simple defenses. Don't ignore warnings. Make sure that helicopters take off immediately.
Instead, they went with מָוֶת לָעֲרָבִים
I assume that the reason Israel did not take precautions was that they did not care. They knew that if Hamas tried a large scale attack, they'd respond with genocidal destruction, which would make whatever government would carry it out popular with the Israeli electorate.
Has Hamas been destroyed yet? No, they're still there. What is stopping them from doing another October 7 tomorrow? The fact that Israel keeps an eye on and defends the border. That is the active ingredient that prevents another October 7. Not genocidal fury, but alertness. And you know it's true because nobody thinks another October 7 is possible tomorrow, or back in 2023 on October 8 or 9 or 10.
Genociding Gaza has cost Israel well over 100 billion dollars. With one tenth of that money you could have built defenses that Hamas would never be able to get past. For the security of Israel, each dollar spent setting up landmines and concertina wire throughout a Korea-style corridor around Gaza would be worth, I don't know, 10, 100, 1000 dollars spent genociding Gaza. Not that Netanyahu cares. His goal is not the security of Israel, his goal is to satisfy the bloodlust of his electorate, and to make it possible for settlers to steal land.
"Calling the Gaza campaign genocide is an ugly smear"
"It was time for Gaza to go"
yeah, it was genocide, you just support what was done to those poor people
And Israel did not try appeasement, what it did beforehand was a perpetual siege with occasional October 7th style murder raids. They never offered real independence or control of their own borders.
Anyone with any understanding of military reality knows that if Israel wanted to kill everyone in Gaza, they could get most of it done in an afternoon.
Instead they fought an urban war for 2 years and only killed 3% of the population.
The genocide accusations can only come from deep ignorance and/or passionate antisemitism.
I support dismantling terrorist infrastructure.
America is not the world police. Terrible things happen all over the globe. There is nothing particular about Iran that makes it morally righteous for us to get involved. Iran is a belligerent, but they are a threat primarily to Israel and other countries in the region. They do not have the technological or military prowess to pose a threat to anyone living in the geographical United States. I am excluding military bases from that calculus, though those are treated as being US soil. Almost every single American casualty is due to us having a military presence in the region.
When you get right down to it, this is about our ally Israel. Iran is not shy about it's desire to see Israel gone. We are supporting them in ending a threat to their existence. When it comes to America, they are an old man yelling at a cloud. But their cruise missiles can physically reach Israel. They bankroll terrorists that kill Israelis. I am not opposed to the air strikes, but I would like to see it end there. This is Israel's problem, not ours.
Right! Why can't we control Israel? We send them enough money.
"our ally Israel"
And our ally KSA
And our ally UAE
And our ally Qatar
And our ally Kuwait
And our ally Bahrein