Re Argentinian economy, there is a difference between Dollarized and Fully Dollarize where they try to Dollarize but it doesn't quite take and they end up with a mix of dollars and the old currency (pesos)
Manifold wants the dollar de jure in place, Metaculus wants the peso de facto gone
Happy to. You could also interpret the spread as "Milei tries to Dollarize Argentina but it doesn't work and/or he is replaced by somebody who rolls it back", which I think is quite possible
A currency peg. They work as long as the economy is stable and the government is trusted, because they rely on the central bank buying and selling currencies to keep the actual rate equal to the posted rate, if they fail to do so there will eventually be a black market at the actual exchange rate and the peg will fail.
Also, smacking Hollywood by bypassing them to put her concert film in theaters and grabbing some of the NFL's spotlight makes it a better year for her.
Also, she has reached the level of being called out by name in a regional Federal Reserve Bank report…
«Despite the slowing recovery in tourism in the region overall, one contact highlighted that May was the strongest month for hotel revenue in Philadelphia since the onset of the pandemic, in large part due to an influx of guests for the Taylor Swift concerts in the city»
Something has definitely happened to Taylor Swift over the past year that has taken her from being part of the cultural background noise, easily ignored, to a huge player that you have to work hard to ignore. Check Google Trends if you don't believe me, there was a spike last October, and then a steady climb from March 2023 to current levels where she's a more popular search term worldwide than "sushi".
I can only attribute this to the fact that (a) she's touring and (b) she apparently has the best PR people in the world. At some point earlier this year I noticed that articles on totally unrelated subjects started talking about Taylor Swift for some reason, and it's only got worse since then.
I don't know how much of it to believe... it almost seems like a fanfiction of reality, but the Kerfuffle stuff is too notorious for this story to have been invented whole cloth, i think...
but gosh, what a read. thank you so much for the link
If you enjoyed that one, don't miss his previous exploit, called "Live from the hate march", on the same substack.
I appreciate the expression "fanfiction of reality". I can almost see the snarky headline on some normie blog, "Internet rationalists discover the existence of literature - now called fanfiction of reality".
When I was reading it, it tripped some memories that I had of something that seemed like it might fit the bill, something about an indie zine owner getting in huge trouble, but in retrospect I might have been Mandela Effecting myself
tl;dr: Swift never sexualized herself unlike Britney Spears, Katy Perry, or Miley Cyrus.
He then goes off on a bizarre tangent about Taylor Swift's navel and how he thought she was Lilith.
IMHO:
I actually find the 'never showed off that much skin' theory credible: mild feminism and attractiveness without excessive sexualization would allow her to appeal to the center-left and center-right respectively, creating a bipartisan audience that rarely exists anymore. But *that* spin is mine.
Hanania had a few posts about pop music that were interesting to me as someone who stopped following pop music around the same time as him (early 00s).
He argued that white female pop stars have for the most part aggressively *de-sexualized* (i.e. androgenized) themselves while at the same time not really singing what we would recognize as a "love song" anymore.
To the extent this is true (and I'm too ignorant of pop music to really know), we might say Taylor Swift is the last pop star. She sings romantic love songs (and break-up songs, but ones that still hold romance and love in high regard) while looking pretty but not pornographic, and she does it highly competently. There used to be a lot of competition in this space, but perhaps now there's basically none?
I think I read that. I don't know enough about pop music anymore to say. Sounds credible. Hanania's a smart guy, even if he says weird stuff sometimes. Maybe she identified an underserved market and went for it.
I'm skeptical on the last sentence. I think it's exceptionally rare for musicians to pivot hard in a musical direction that goes against their instincts in pursuit of commercial success, and to then succeed for any length of time. I'm aware of two credible alleged examples: the bands Bush and Sugar Ray. In both cases, they didn't really last long.
There's a documentary about Kenny G, and it's funny because they interview all these "jazz experts" and they all hold him in contempt, describing him as a complete sellout who saw the market opportunity from watering down jazz music to make it palatable to the masses. And they interview Kenny and he basically says, "I'm really not that smart. I just make the kind of music I like to make." I believe him.
No. This is completely wrong. Billie Eilish, Olivia Rodrigo, Ariana Grande, I can keep listing names but I can say if your summation is correct Hanania is ludicrously wrong again. Like, the main most popular new white female pop stars still sing mostly about romance and look conventionally attractive.
Billie Eilish - Richard describes her as "going out of her way to look less pretty than she otherwise could" and based on an image search this seems apt to me. I also didn't think she really sung about romance; based on her debut album cover art and some vague things I heard, I thought she was positioning herself more like a young female Marilyn Manson, but I suppose I'm wrong there. But I look at her album tracklist (and again, that cover art) and see titles like "all the good girls go to hell" and "wish you were gay" and we're clearly in very different territory from Taylor or any conventional romantic songwriter of the past.
Ariana Grande - Maybe she bounces around stylistically, but I Googled her top songs and the first one that popped up was called "7 Rings". If you look at the lyrics and music video, it clearly falls into the "bad bitch" style that Richard describes.
Olivia Rodrigo - I hadn't heard of her, maybe a more promising example than the others. Though she's apparently half-Filipino, doesn't qualify as "white" in the US.
EDIT: Just looked at Richard's piece again and he DID mention Rodrigo as an example of someone who produces a new-style breakup song that is indicative of helplessness and possible mental illness, as opposed to the breakup songs of the past. Or, we might say, those of Taylor.
Yep. I don't actively look for anything TSwift related, but it seems like, in terms of career success, she's gone from "probably the most successful female artist of this generation" to "almost certainly a top 2 most successful artist ever" (alongside The Beatles).
(Also, Scott says, "she's been successful for the past 10 years as well" but frankly she's only become more successful over time *and* entirely changed genres, despite already being quite popular in 2008, which is actually quite rare among musicians. Who else was big in 2010? What are they doing now?)
As for albums sold, Chartmaster puts Swift in 16th place. Above Mariah Carey, but below Rod Stewart. However she's moving up fast, and it's expected she'll knock Celine Dione out of 14th place soon. The only solo female artist above Dione and Swift is Madonna. She's in 5th place.
I'm not convinced lists like this do a good job of accounting for changes in how music is consumed over time. I know they make an attempt, but even the "selling" in the phrase "best selling artist" isn't really accurate when so much is available for free. I'm also not convinced that the way they account for international developments is very good, since they rely on a separate organization in each country, each of which might do things differently.
Different lists aren't even consistent. For example https://en.wikipedia.org/wiki/List_of_best-selling_music_artists has Swift in 9th (counting by "total certified units) with Madonna (who's been active for far longer) and Rihanna above for female artists, but this metric would have Rihanna as by far the most successful artist ever. Mainly I wanted to link it because it also gives numbers for "claimed sales" which vary wildly within each artist (when multiple numbers are given), differ dramatically from "total certified units", and are only presented with very approximate values, highlighting the difficulty of measurement in this category.
I agree with you for the most part. Also, remember that this definitely doesn't reflect the size of the fanbase of any given artist at the time of this writing (2023). Many Elvis fans are now passing away. And Beatles fans are in aging and dying too. I doubt that either the Elvis or the Beatles are enlarging their fan base faster than their fans are dying off. Heck, it's the 40th Anniversary of Michael Jackson's Thriller! His fanbase must be on a general downward trend now.
Using Google's ngram viewer, it looks like the Beatles will drop from cultural consciousness sometime between 2070 and 2080, and Elvis, though he peaked higher than the Beatles is dropping faster and will probably be extinguished from our cultural consciousness in by 2060. Unfortunately, ngram viewer only tracks word frequencies up to 2019, so we cannot really see where she stands today against other pop music icons.
Heh. Every generation rediscovers the power of branding, and is surprised at how powerful an effect it has on the popular culture. She may be this generation's Oprah.
Yep, she's exploded this year. There is a reporter dedicated to covering her, Outlets are discussing how much GDP she produces while in your, Freddie and others have started ranting on how she and others cannot be criticized. My Father in his 60s has started calling himself a Swifie and gushes over how cute a couple she makes with the NFL star. She's the biggest pop star since Michael Jackson, who was also a Time Person of the Year.
She's basically at Madonna or MJ level now. Before this year I sorta lumped her into a group with e.g. Lady Gaga and Katy Perry, as female pop singers who I knew were famous but I rarely see in the wild due to my apathy for Top40 music, broadcast TV and celebrity gossip. But she's clearly peaked well above the level of other similar artists, for whatever reason.
Re-recording several albums is nutso, shouldn't have worked, and definitely couldn't have worked in prior eras of music, but apparently did work due to harnessing absurd levels of fan loyalty and manipulating the streaming services. The well-timed fake romance with Travis Kelce, where everyone very obviously avoided saying anything untrue while still giving the impression they were dating, had to involve tons of media people playing along, but they did since it was in their own self-interest to do so. I don't know how much she herself was a driver of these recent moves, but it was rather clever. I think MJ's music was more ubiquitous, I still can't name more than 3 of Swift's songs, but in a fractured media landscape where she can't put her videos on Fox before the Simpsons and draw half the TVs in America, she's gotta be close to the peak stardom possible.
> Re-recording several albums is nutso, shouldn't have worked, and definitely couldn't have worked in prior eras of music
What about it is nutso, or unusual?
Seems to me that most performers routinely re-record their studio releases by doing live performances. A live album is mostly just re-recordings of earlier work.
Re-recording studio versions of an album and then selling it again is unheard of. Live versions are quite different for rock acts, maybe less so for pop acts but still offering something distinct.
This as I understand it was done only to own the songs, and clearly depends on Spotify and iTunes etc to offer the new version without any artistic justification for it. Whatever company now owns the original “1989” and other records can still license and sell the music, so the entire thing relies on Swift convincing people only to stream and purchase the version she owns, out of fan loyalty or antipathy to big business (as if she herself wasn’t basically just a giant corporate blob at this point.)
My understanding is that TS still holds the songwriter copyright, and is using that to prevent the original catalogue from being played. The story I heard is that her original recording contract held the copyright for the recordings. The label sold them to someone (not sure of what the history there is, but my understanding is playing TS's "Bad Blood" will provide proper mood) for $900,000,000. Of which TS received $0. So withholding songwriter permission for performances lets TS zero out the value of that property, while recording the "Taylor Version" gives her a steady income stream.
Wasn't the label itself sold to someone Swift has accused of bullying? And possibly for mere $300 000 000 – $400 000 000.
Apparently Swift cannot block exploitation of the old recording via copyright means. She seems to have enough consumer loyalty to cause a consumer boycott of those old versions though.
> So withholding songwriter permission for performances lets TS zero out the value of that property, while recording the "Taylor Version" gives her a steady income stream.
(EDIT - this following paragraph is wrong; Taylor Swift herself always held the publishing rights to her songs, and the royalties owed on the covers are therefore owed to her. That said, the value of the originals is clearly very high, not zero.)
But the Taylor's Version raises the value of the original back above zero, because she has to pay royalties when her covers play.
Others have re-recorded their albums (and/or just individual songs), sometimes for basically the same reasons, other times because they wanted to do it differently or better.
In most of those cases it does appear there was an artistic reason cited or an obvious stylistic change. I still think it's unique to the Swift situation that she was both able to re-record an entire back catalog for purely financial reasons AND that the current music distribution landscape is such that this maneuver actually works. The Everly Brothers and Sinatra examples are cases where they re-recorded for new compilations, which makes sense in the days of physical media that you had to be offering a new product to sell. But when everybody just streams music, they're accessing it anew each time they want to listen rather than keeping a pre-existing already purchased music library, so she could lean on Spotify etc to replace the old versions with the ones she owned.
As an analogy, imagine Garth Brooks had tried this 20 years ago with his first 7 albums. There were already too many copies of those CDs in his fans' hands to get them to re-buy all of them, best he could've done was slap them all into a box set and hope to sell enough to fans that didn't already own them, but that wouldn't have justified the expense. Only in the current age, and only if you can implicitly threaten to unleash your army of drone fans like Swift can, does any of this work.
A live performance is very much not the same thing as a re-recording; recording generally involves multiple takes and layered parts. Performing an album live takes as long as the album runs. Re-recording an album can take months.
If one considers amount of news coverage, I definitely consider her in the running. By that metric, considering the competition by year she's definitely had a bigger year than previously.
Swift has retrieved the rights to a bunch of her old songs, re-recorded them and released new versions as albums. She is also touring extensively this year.
She was well-known and famous before, but these factors have combined into some sort of next-level fame.
> Swift has retrieved the rights to a bunch of her old songs, re-recorded them and released new versions as albums.
I thought the point of re-recording them was that she didn't have the rights to the old songs? If she owned the old recordings, what would be the purpose of re-recording them?
I guess the reason for the new fame that the fans are taking her side on the dispute, which is why they're buying the new versions in more than the usual numbers that would happen for e.g. a remix album.
I also note her dad is an ex-stockbroker and her mom was a mutual fund executive before being a homemaker, so evidently she has genes for business. Her parents may have taught her something too.
> but recovered rights to make new recordings of the same songs.
Huh? Everyone has the right to cover a song. It isn't possible not to have that right. You can't "recover" what you never lost.
You need a license to record the cover, but issuance of the license is mandatory (and doesn't involve talking to the copyright holder) and the fee is set by law.
Yeah, I might look dumb in a few weeks for saying this, but the fact that the prediction markets are giving Altman a higher chance of being POTY than Swift doesn't speak well of them, imo.
Semi-snarky nit: Should ChatGPT really be in the running for _person_ of the year? This feels like jumping the gun on claiming to have AGI to me... Or does passing the Turing test (with whom as the judge?) suffice?
Well done, though I don't think this was as easy a call as you say. They've never had a pop star or for that matter any kind of entertainer as person of the year. They've almost all been some sort of political, social, religious or business figures.
Yeah, I would've been less confident if I'd realized an entertainer had never won it, but 1. I think she was clearly the most likely winner on the shortlist (Hollywood strikers, Xi Jinping, TSwift, Altman, Trump Prosecutors, Barbie, Putin, Jerome Powell), and 2. Having Altman at twice her price was just insane, so it was pretty clear to me she was undervalued. Time is in the business of selling magazines, and if they'd put Sam Altman on the cover, 90% of people would've said, "Who?"
Interesting how an entire thread on the world's biggest pop star at the moment doesn't seem to include a single word about whether her music is any good. I take it as an indication it might be not - but maybe I'll have to do the supreme effort of firing up youtube and checking for myself
OK, had a good listen around her albums. Not exactly my cup of chai, but she's clearly a good artist. Excellent production, good songwriting, expressive voice, good grasp of the zeitgeist, variety of styles, and weaving a fine line between emotionally engaging, assertive but also kinda harmless. Lots of moody minor chords, tonal ambiguity and electronic textures. "Red" is probably the best of what I've sampled - but I still prefer the King Crimson album of the same name.
Ha! I fear that opinions about the quality of her music will be more subjective than discussing stats about her reach. Personally I’m fairly indifferent about her, but she certainly seems to deeply resonate with a large number of people. I suspect it has something to do with the fact that she writes her own songs - not that common in pop world - and she combines her talent with a vulnerability and openness that seems to connect with fans who like to dissect the meanings of her lyrics (even if it’s as apparently shallow as “which ex is this one about?”) and they end up feeling like they know her deeply or something like that. Rumor has it she’s also got some business savvy.
The Eras Tour was the highest grossing tour in U.S. history. It generated small booms in excess consumer spending in every city where she performed; According to current estimates, about $4.6 billion in excess consumer spending. It's like Swift ran her own private stimulus program for post-COVID downtowns. If you worked in or near a city she performed in it was all people were talking about for weeks. This is pretty much the largest cultural phenomenon in the US in years and years.
How is a commitment to merge with other labs close to superintelligence even plausibly a good thing to commit to if you are concerned about alignment? What if those other labs have different ideas about alignment? What if one of them actually wants an unaligned AI?
I mean either it's essentially meaningless or it's dangerous. If it only requires you merge if the other labs acceed to putting you in charge or at least adopting the approach you prefer then it's just kinda silly. I think pretty much any CEO and board of a non-profit will always accept some other organization with a bunch of assets showing up and saying: you can direct all our resources too.
OTOH if it means serious negotions to work out a power sharing agreement that induces other institutions to accept then it obligates you to share power with groups whose goals (or at least methods) are directly opposed to your own. And if the then board gets to decide if any other institution is too incompatible it's really saying: merge if it seems wise to do so which is the same as not saying anything.
--
My concern isn't the legal enforceability. Often it's good to put in commitments to a charter that lack any realistic enforcement mechanism. Rather, it's that either good faith board members are either agreeing to a rule they should know would be quite harmful if they felt obliged by it or it's misleading to phrase it as a commitment and should really be more like: when you get close stop and do whatever helps.
You're close to ASI, and have competitors who are also closing in. There is the fog of war. Would you prefer to rush alignment so you can be first (increasing risk), or try and negotiate with the other labs so all of you can actually work out alignment?
The charter says "if a value-aligned, safety-conscious project comes close to building AGI before we do, we commit to stop competing with and start assisting this project." I think the theory at the time was that a major issue wouldn't just be solving alignment, it would be making sure that the first group to ASI would have enough of a lead to be careful about implementing it, and wouldn't rush the last steps because of uncertainty about less-careful competitors.
Ok, thank you. That explains the thought process. I was just thinking about it quite differently.
It does basically commit openAI to scuttle any attempt at a pause and seems to ignore the possibility that the first group to get close might be motivated by other concerns than capitalistic ones -- eg it won't be people who are pursuing the project because they believe it will save their country from some kind of war or political crisis.
> a graph of different people’s probability estimates of AI causing human extinction
Are these probabilities of extinction conditional on AGI or extinction in general? There's plenty of future outcomes where AGI is never built for one reason or another.
What are some of those? Absent AGI, or similarly advanced tech, there are no plausible total extinction scenarios in the foreseeable future. Technological progress is a robust attractor that civilization would eventually converge to even if the current iteration suffers some catastrophic disruption like a global thermonuclear war, which would once again put it on the path to AGI, another robust attractor, and seemingly in easier reach than Drexlerian nanotechnology or brain emulation.
You don't need full nanotech to end the world; an alga that can't be digested, has a better alternative to RuBisCO and needs 1000x less phosphate is enough (because it blooms like crazy across the ocean and ties up all the atmospheric and then biospheric carbon as gunk on the seafloor). This is *mostly* a computational problem at this point, although the assembly would also be mildly beyond current SotA.
Redirecting a sufficiently-large asteroid into Earth would do the trick if self-sustaining offworld colonies did not yet exist. Chicxulub wouldn't be enough, Ceres would, not sure about the actual crossover point.
Scaling up particle colliders enough could lead to micro black holes.
The amount of dakka required for these is large, but they don't really need super-tech.
All of those seem to be deliberately self-destructive while requiring non-trivial effort, which, to me, puts them below the level of consideration. After all, we could always have just murdered everyone with sticks and stones, but somehow have managed to avoid doing that for millions of years. Meanwhile, stuff like AGI and full nanotech are dangerous precisely because of their vast non-destructive potential, but which could catastrophically spiral out of control.
I agree that in general it's not worth giving these scenarios much consideration, but I don't think I agree with your reasoning. Individual human beings do, indeed, murder people with sticks and stones. They're a small proportion of the population, though, and they can't kill more than a handful solo before someone stops them. the probability of so many individual humans all committing a bunch of murders all at once is vanishingly small.
I have no doubt that there are individual human beings who really existed who, if given the opportunity, would have killed more but lacked the capability to do so.
once nukes were on the scene, the bar for human-caused extinction finally came within line of sight of our species, but even getting close still required the resources of a major national government. as the bar lowers, the number of simultaneous crazy-psycho events needed to seriously screw things up drops
as technology advances, that bar gets lower, eventually it gets so low that, idk, a random dealthcult with a single microbiology PhD could pull it off. if we survive that, eventually it gets so low that a single person with a garage bioreactor could do it.
I would not be so quick to write off 'obvious malicious self-destruction' timelines
>Individual human beings do, indeed, murder people with sticks and stones.
Overwhelmingly pursuing their self-interest. There were a few moderately "successful" suicide cults over the centuries, but I wouldn't expect their future analogues to get world-destructive capabilities before actually powerful actors are in position to destroy the world seeking beneficial ends.
The first one will be sold as «highly efficient biological carbon capture». Won't even be a lie. No need to reveal (or notice in time) that your breakthrough technology just happens to have side effects.
The particle collider and the Life 2.0 can happen without actual malice; we have GoF research right now despite it being -EV and despite the people doing it not generally wanting to kill millions with a pandemic.
The asteroid does need malice, but it's plausible that someone living off-world might not understand that his/her colony is not self-sufficient and thus it only absolutely requires war malice, not omnicide malice.
I do agree that the continued GoF clown show is some evidence that humanity could sleepwalk into self-destruction without actual malice, but I'd still in general expect several progressively worse non-total-extinction warning shots due to such brainfarts.
Pandemics and asteroids yes, Life 2.0 and particle collider no. A particle collider either produces ice-nine of some sort (black holes, stable strangelets, true vacuum, etc.) or it doesn't; there's no in-between. Life 2.0's not quite as obvious - bad-but-not-X Life 2.0 is physically possible - but the things that would kill us are almost the easiest things to get working at all.
WWIII isn't an X-risk. GCR, definitely, but not X. Direct blast/heat/etc. can't kill rural communities (or towns/cities not nuked, and there aren't that many nukes anymore), fallout decays to tolerable levels in weeks, and nuclear winter a) is mostly a hoax, b) would have to be hilariously, unrealistically bad to actually finish us all (as even hunting remaining possible would preserve the species, so the entire ocean would need to freeze over).
Both biotech and nanotech are possible X-risks. Diseases are not designed to kill people but if the technology for designing them gets good enough they could be. Grey Goo scenarios with Drexlerian nanotech are possible. Either could conceivably wipe out our species.
It seems like these AI experts have a blind spot in that they are unable to consider the likelihood that AI could be beneficial to humanity. When discussing a range of probabilities, if there's a chance that AI could cause human extinction, what's the chance that the AI could be extinction-neutral? — or could *prevent* human extinction? I can just as easily imagine a Banksian Culture-like future where AI 'Minds' have endearing and eccentric personalities and get along well with humans.
I think there is a deep psychological need in many humans to indulge in apocalyptic fantasies. Since the 1970s I've been told by "experts" that humans are facing extinction by a continuously-changing set of threats—
1. Pollution (will kill the O2-producing algae)
2. Club of Rome prediction that civilization will collapse by the year 2000
3. Global cooling and the coming ice age
4. Nuclear winter
5. AIDs will infect one-third of humanity (ending civilization);
6. Global warming
7. Ocean acidification (will kill the O2-producing algae)
8. Asteroids
9. Y2K
10. COVID-19 will be a mass disabling event (ending civilization)
11. and now it's AI.
Those first five predictions haunted my childhood and young adulthood until I started to realize that most experts were either purposely bullshitting us and/or themselves.
Meanwhile, my Xtian neighbors kept warning me that the End Times were coming any day now!
Neither the experts nor the religious have been correct in their predictions. I just take them all with a grain of salt now. Humans are incapable of predicting their future, and it will most likely be something we didn't consider that will take us down. ;-)
Ah, you remember that too? A good while back I got plenty of "nope, never happened, nobody ever said that, you're mistaken/lying" from the disgruntled CLIMATE CHANGE IS GONNA KILL US ALL types on here when I mentioned that the coming New Ice Age was one of the flaps of my youth.
At least if I hallucinated it, I'm not alone 😀
Oh, and don't forget: hole in the ozone layer! We were all going to get skin cancer and die from that one.
Eh, to be fair, most of these are attributable to the journalists (who have only gotten worse lately), not to the actual scientists (nor even policymakers).
1. "Pollution (will kill the O2-producing algae)": Still a real problem, only it depends on the kind of pollution. For example, fertilizer pollution creates delightful and vibrant algae blooms (that are also toxic to all other forms of life).
2. "Club of Rome prediction that civilization will collapse by the year 2000": No idea what this is, never heard of it. Sounds like a newspaper thing.
3. "Global cooling and the coming ice age": Was never a real thing, media companies invented this one.
4. "Nuclear winter": And that's the least of your problems if global thermonuclear war actually happens; and with events like the Cuban missile crisis, it seemed like a real possibility... and it still is to some extent. But for now we seem to have pulled back from the brink.
5. "AIDS will infect one-third of humanity (ending civilization)": It's kind of already the case in Africa, sadly; but the world-wide prediction was a media invention.
6. "Global warming": Would be lovely to say it's all fake, but right now looks like it's all too real. Climate skeptics have mostly switched from saying "it's a hoax" to "it's totally happening but maybe that's for the best". Hmm.
7. "Ocean acidification (will kill the O2-producing algae)": Not the algae (they have a pretty quick evolutionary cycle), but certainly the coral reefs. If you wanted to see one in all its glory, now's probably your last chance.
8. "Asteroids": They're a real though relatively unlikely threat. Have you seen the Chicxulub crater ?
9. "Y2K": This one was a combination of real problem that was successfully solved by a concerted effort, and OMG media overhype to the max. If we did nothing, chances are that e.g. banking systems would crash, which would be... not great.
10. "COVID-19": Could've been much worse without MRNA vaccines, but like AIDS, somewhat overhyped.
11. "AI": What pisses me off here is that all the Singularity hype is masking the real danger. There are people now who are cheerfully letting ChatGPT write their laws or even their software. This is like putting a typewriter monkey in charge of same.
12: "Ozone Layer": This was a very credible threat which was solved through a concerted effort (banning CFCs). For once, humans did something right... sadly we probably lost the capacity for such action by now.
"Concern peaked in the early 1970s, though "the possibility of anthropogenic warming dominated the peer-reviewed literature even then" (a cooling period began in 1945, and two decades of a cooling trend suggested a trough had been reached after several decades of warming). This peaking concern is partially attributable to the fact much less was then known about world climate and causes of ice ages. Climate scientists were aware that predictions based on this trend were not possible - because the trend was poorly studied and not understood (for example see reference[16]). Despite that, in the popular press the possibility of cooling was reported generally without the caveats present in the scientific reports, and "unusually severe winters in Asia and parts of North America in 1972 and 1973 ... pushed the issue into the public consciousness".
Well, tell that to my (late) Glacial and Quarternary Geology professor. He had graphs and data showing that we were entering the cooling phase of the current Milankovitch Cycle. It was taken very seriously despite what Wikipedia might say. Suggest at John Imbrie's work from the 1970s and 80s.
"2. "Club of Rome prediction that civilization will collapse by the year 2000": No idea what this is, never heard of it. Sounds like a newspaper thing."
"In this context climate models offer a useful point of reference. General circulation models for the atmosphere and the oceans, along with related models of ice sheets and atmospheric chemistry, have several points of similarity with World3. At a conceptual level the structure is much the same: There are flows of air, water, heat and other entities, which the model must sum or integrate. The time scales are similar: In both cases we want to know what’s going to happen several decades out. And feedback loops are essential mechanisms in both kinds of models. (There are even historical connections. The use of general circulation models to study global climate change began in earnest at MIT circa 1970. The instigator was Carroll Wilson of the Sloan School of Management, who was also the person who got Forrester involved with the Club of Rome.)
These similarities are outweighed by differences. Where the Limits team had a casual attitude to data gathering practices—and outright hostility to statistical methods—the climate science community is passionate about collecting data, verifying its provenance and quantifying its uncertainty. General circulation models are not based on rough estimates or guesses but on decades of meticulously curated measurements—what Paul Edwards, in A Vast Machine , calls a “climate knowledge infrastructure.” "
The other big difference is that _Limits to Growth_ ignored rational human responses. In their model, if food prices went up, output went up in the short run then down as farmers destroyed their soil trying to produce more. That assumed that the farmers ignored the fact that high prices gave them an incentive to maintain the soil for future output. As I put it at the time, the model was like predicting the path of an automobile whose driver couldn't see through the windshield.
Climate science doesn't have that problem as long as it is only dealing with climate. It only comes up when it is being used to predict the effects of climate change, which isn't really climate science. The amount of CO2 depends on human action but the ICCP can and does deal with that problem by considering multiple emission scenarios.
1. Ocean anoxic events (OAEs) have happened in the past (probably significantly to the Permian mass extinctions), but the claim that pollution and/or acidification will cause an OAE event seems pretty far out on the range of probabilities. Especially since we've significantly cut down on our water-polluting ways since the 1970s—and because the evolution of plankton during the Jurassic that sequestered carbon in calciferous sediments will buffer acidification in the longterm.
2. Club of Rome and the famous/infamous publication "The Limits to Growth". ITheir report kicked off a whole slew of books about the uncertain future of humanity...
3. Global Cooling was a real worry. If glaciation is dictated by Milankovitch cycles, then we're entering a period of global cooling. See the papers by John Imbrie.
4. Carl Sagan and nuclear winter. Considering there were over 500 above-ground nuclear tests during the 1950s, do we have any data that suggests that they affected the climate? I don't know. I haven't looked into that question.
5. "AIDS will infect one-third of humanity (ending civilization)". Can't find the links because they're pre-Internet, but I think Douglas Feldman was sounding the alarums about extremely high mortality rates from AIDs. This was before they really had a handle on its R0, though. I remember Scientific American amplified his concerns (fears).
6. I'm afraid I'm pro-global warming. Until recently CO2 levels were the lowest they've been since the Permian. Angiosperms—which are responsible for most of our food supply—evolved when CO2 levels were 4x-5x higher than they are today. We can see a positive response to higher CO2 levels to the current greening of the world.
Skip to 12. I hate to tell you but the Ozone hole is still opening up every year over the Antarctic. 2023's ozone hole briefly opened enough to be 12th largest hole since they started monitoring it—but overall, 2016 was the 16th worst from Sept. 7 to Oct. 13...
How could I have forgotten about the Ozone Layer?! There's still a big hole that opens up over the South Pole every year, It was the 12th largest on record in 2023.
I was taking classes in Glacial and Quaternary Geology when the narrative transitioned from a consensus of global cooling to a consensus of global warming. It happened within a few months by a few high-profile think pieces in the Scientific American and Science magazine (and maybe Nature, but I couldn't afford to subscribe to Nature at the time).
This documentary has survived the cultural memory loss...
> It seems like these AI experts have a blind spot in that they are unable to consider the likelihood that AI could be beneficial to humanity. When discussing a range of probabilities, if there's a chance that AI could cause human extinction, what's the chance that the AI could be extinction-neutral?
Isn't the answer, according to the chart, that AI experts think AI will most likely be beneficial or extinction neutral? There are only a few people there who put the odds of extinction from AI above 50%
I think there are a couple positions on how to balance AI benefits vs. risks:
1. AI would be incredibly beneficial to humanity if we got it right, but there's no chance we will, so let's focus on not dying.
2. AI would be incredibly beneficial to humanity if we got it right, but better to take time to do it right and wait an extra 10 years for the benefits than eat a 20% risk of dying.
3. AI would be incredibly beneficial to humanity if we got it right, but we should also work on controlling it, for the same reason that even though planes are beneficial, we work on ways to prevent them from crashing.
Given that people have written whole books on these things and balancing the benefits and risks, I think it's unfair to call it a "blind spot".
Also, you're doing a thing I should give a fallacy name to and write a post about, which is giving a long list of things that didn't happen, then using it to prove nothing ever happens, without also including a list of things that did happen:
- The Black Plague
- The Mongols killing 10% of the human population
- Two World Wars
- Global warming of the type we've actually gotten
- AIDS killed however many people it actually killed
- COVID killing however many people it actually killed
Before all these things (except maybe the Plague and the Mongols, I don't know enough history to know what people were saying there), people were saying it was alarmist and would never happen, and those people were wrong too. I think the probability (something bad happens|people predicted something bad would happen) is moderately high and definitely not zero.
I admit humankind has never been destroyed before, but that's a tautology - it's logically necessary that the day before humankind is destroyed, humankind will never have been destroyed before!
> I think there are a couple positions on how to balance AI benefits vs. risks...
As usual, I'd like to point out that alternative positions do exist, i.e.
4. AI could potentially be incredibly beneficial (or possibly harmful) to humanity, but its current iterations fall laughably short of that goal, and there is as of yet no clear path from here to there. We should continue working on AI research, but meanwhile let's make sure we use existing (and incredibly unintelligent) machine learning systems responsibly and without undue exuberance.
a) Re 8. Asteroids was always a low-probability-per-year hazard. I don't _think_ I ever heard it painted as a will-kill-us-in-the-next-decade kind of risk. ( The Yellowstone caldera giving us a supervolcano eruption is similar)
b) For a _lot_ of these, there is an immense amount of noise and oversimplification injected between what researchers in the field concluded and what the pundits spouted
This is late, but for your reference, up until 5-10 years ago, I was largely pessimistic about the long-term future, because I thought we were stuck on a single planet and there were too many things we could do to end our civilization or existence. My hopes were placed in the creation of self-sustaining, self-replicating space colonies, or building something smarter than us to solve the problems we can't. But this was all centuries or millennia away, and didn't affect my day-to-day thinking.
On the first angle, Elon Musk happened, and despite his erratic behavior, I'm quite happy that we're finally moving in a positive direction again.
On the second angle, it's true that hanging out in Eliezer-Yudkowsky-adjacent spaces got me thinking about AI risk in a more serious way. But the main cause of my pessimism came from a close encounter with a malevolent human with superior social skills. The aftermath convinced me that practically no one thinks about that sort of risk clearly, with the notable exception of Eliezer-influenced folks. The only reason my p(extinction) isn't higher, is that it's blocked by a large p(total civilizational collapse) before an AI gains the capability to kill us all.
Hey, MY hair is that stylish and I’ve been using Manifold for over a year! Deleted my manifold.love profile because of privacy concerns, but if you want IRL proof re: my hair, text your shidduch resume to 234-5[gematria of my first name][taryag].
> I’m most unsure about whether Adam D’Angelo is a committed safetyist. He hasn’t said so explicitly and isn’t openly linked to the safetyist movement.
He's not particularly concerned about existential risk, especially in the near future.
Speaking of this guy, does anyone have a bit of insight on what he's even doing on the board of what is arguably the world's hottest company? As far as I can tell he's mostly known for being the CEO of Quora, a much hated site that basically SEO-spams Google with mostly crap user-generated questions and answers. I'm sure the operation makes plenty of money, but still... what gives?
I heard somewhere that putting him on the board was part of a deal where openAI got to train on Quora data, which is copious and already ranked by answer quality. Also imo lots of quora answers are pretty good.
I must not be in the target audience... most of the time I find much better info on wikipedia and random blogs, and even by asking chatgpt, despite its hallucinatory habit. And I can't stand login walls - we live in an information glut, if I have to jump through hoops to read your content, I just close the tab and look elsewhere.
Scott copies the usage in his first paragraph discussing it, but then switches to "oust" in the paragraph following that.
It's close enough to making sense that I'm not really comfortable declaring that it must have been an error by the question author and then another error by Scott, but I can't rule it out either.
I think it's an error by Scott, influenced by the question author's error. But to answer your original question: no, that's not a standard usage for me either, to the extent that I assumed it had to be an error.
It's weird that the only time Lawrence Summers had anything to do with tech companies is having one meeting with the Winklevoss twins vs FB during his Harvard years. Everything else is politics, World Bank, Harvard, hedge funds and the Treasury/National Economic Council stuff.
He has, however, mentioned AI before:
> In 2018, Summers disputed the claims from then Treasury Secretary Steve Mnuchin that AI would not replace American jobs for 50 to 100 years.
> “The robots are coming," Summers wrote in the Washington Post. That year, he also warned of economic catastrophe if the US “loses its lead" in biotech and AI to China.
It's not helpful to think of cancelation attempts as a binary, because Summers seems to have landed somewhere in the middle of the spectrum. Wikipedia cites this commentary on the end of his Harvard presidency and on his less prestigious role in the Obama administration (compared to the Clinton administration): https://www.independent.co.uk/news/world/americas/summers-sexism-costs-him-top-treasury-job-1033373.html
Sure, but he's already been cancelled once, which means he's more likely and able to stand up to cancellation threats in the future. That may or may not be part of their calculus.
I wonder if you could make it easier for prediction markets to handle "why" questions by mixing them with Pol.is-style crowd analysis (incetivize users to submit and vote on binary statements regarding the market->Use a dimensionality reduction algorithm to find the primary dimensions of disagreement->Let the options of the market be defined by these dimensions).
That's cool! But also complicated. I think it would help a lot to just filter submitted answers through a poll of random other users to see if they think it's a good answer (clear, not a repeat, etc) and require 5/10 to say yes.
I wish them luck. It would be neat to see "why" questions usefully examined by prediction markets. Though I shudder to think how difficult and contentious the judging task is going to be...
> Are some of these people normies? Seems surprising, but I can’t rule it out.
I guess it also depends on where you draw the line? For an extreme example, some maths PhDs have many normie leanings and count as normies in some contexts, but are definitely not close to modal normies when the question is about enjoying the precise-sciences in-jokes…
Some of those profile photos are pornbots (as we call 'em over on Tumblr). I image searched a couple because they looked a little bit too identical, and one at least is an image for modelling bikinis, which is also used as a profile photo on an X account that is - I won't say OnlyFans because I don't know, but that kind of "follow me for hot naughty content" bait.
So congratulations, the dating prediction market is mainstream enough to have the bots and spammers interested! 😁
Wait, no. That’s a terrible idea. Scammers could make real money by spending fake money (mana) to short the fake-money price of their “is this person a bot” market. This would cause lonely nerds to lower their prior on that person being a bot, making them more likely to get scammed.
Rough rule of thumb: if the person's photo has them with their bazooms bustin' out of their tops (male or female, equally applies) and pouty vixen/saucy expression, it's a bot. Or at least, a very good chance of same.
Many of the people on those profiles are genuine because they look like people who would go on a Rationalist dating site (I know I'm leaving myself wide open on that one for "So what exactly do you mean by that, huh???" but yeah. You know your people.)
> This will be a total unfair stereotype, and you should feel free to yell at me for it, I just think usually prediction market junkies don’t have hair as stylish as the woman in Picture #3
Ok, I’ll do it- I think this comment is unnecessary and pretty alienating. I understand that it’s lighthearted, but people feel hesitant enough already about not “fitting in” to the rationalist community that and jokes like these reinforce that. Also, putting myself in that person’s shoes- I wouldn’t like to submit a photo to a dating site and have it remarked on to a wide audience in any way, positive or negative.
On the other hand, if you're putting a photo up on a dating site, you are opening yourself up to comments by people at large, even if that's just the people also using the dating site. You will be judged by them on "ooh, love that hairstyle/oh gosh, what a raggedy mop" as to whether they'll engage further with you, and sometimes maybe you need advice as to "hey guys, not getting any matches, what am I doing wrong? here's my profile, take a look" and yes, it's "ditch that photo, it's not doing you any favours".
Yeah, I acknowledge this, but I was trying to make an important point (it seems like Manifold Love has spread to a normier demographic than I expected) and I couldn't think how else to do that. Pointing out the black guy in the hoodie would have been even worse! But I don't think I'm imagining the demographic shift, and I'm not sure how else to evidence it.
Some people have proposed that all of those accounts are spambots because they don't have a profile, but ironically the woman with the stylish hair does have a profile and seems real.
> I wouldn’t like to submit a photo to a dating site and have it remarked on to a wide audience in any way, positive or negative.
This is more or less exactly what pictures on dating sites are for. You’re intentionally broadcasting your photo to a wide audience in the hopes that it’s positively noted. There’s no way one can seriously claim there is a privacy or abuse implication to this.
Also, the implication that this person would be *upset* by having someone note that they have a more stylish appearance than the median person of their in-group, and therefore it would cause them anxiety centered around belonging to the group, is… inexplicable.
Nobody ever got turned away from a lunch table in the cafeteria for being too good-looking.
Having many people separately evaluate your appearance, for the purpose of dating, is different from having someone use your appearance to make a point in a blog post.
Yes, but not in any way that matters. I think you’re having a hard time making a specific criticism because there’s not anything for you to find. This is reaching.
> Nobody ever got turned away from a lunch table in the cafeteria for being too good-looking.
«Stylish» means not just too good-looking, it includes choices beyond taking care of the health side. And «if you come here wearing a suit, you don't really belong here» is not and unheard-of sentiment in tech communities. So «too stylish to belong» is not outlandish enough to be unthinkable as meant literally.
Of course what matters is how the actual phrase written — in the context of the disclaimers, which do exclude literal meaning — reads for people who could be plausibly meant by references to stylishness. And it matters that both «ACX readers» and «Manifold players» are skewed towards having an idea what weak evidence is. I can't be accused of being stylish in anything, so no idea how it works out at the end.
I'm a huge Swiftie, so I don't claim to be unbiased. But the Swift mania is off the charts more than ever this year, and with good reason. Here's the best case I can make for her as Person of the Year:
1. Eras Tour
Everything about the Eras Tour is absurd. Who performs for 3.5 hours straight with 10 different costume changes? And then does that for several nights in a row? And then repeats the spectacle for 146 nights over nearly two years? (The tour started in March 2023, and the international leg continues until December 2024!)
Her opening night brought in more business for the host city than the Super Bowl: https://time.com/6307420/taylor-swift-eras-tour-money-economy/. And that was just one of 52 shows in the US alone. Overall, her tour brought in nearly $1 billion in revenue this year, making Taylor a billionaire. And it produced around $5 billion in revenue for the broader tourism industry.
In addition, the millions of people who attempted to buy Eras Tour tickets crashed the Ticketmaster site so badly that it led to a congressional investigation. This included multiple senators dropping references to Taylor's lyrics in an attempt to seem cool and hip: https://www.cbsnews.com/news/taylor-swift-ticketmaster-senate-judiciary-committee-eras-tour/. (IMO, all of them failed as miserably as one would expect, but the obvious attempts at pandering show how influential Taylor is.)
Taylor's success is due to her business savvy just as much as her songwriting ability. She's constructed a brand around herself in ways that few other artists can touch. And the cornerstone of that brand has been her brilliant moves over the past few years to gain control of her own work.
To recap, in 2019, the original masters of Taylor's first six albums were sold to a private equity firm over her strenuous objections. But Taylor found a loophole - the copyright only applied to those specific recordings. As was standard for the industry, Taylor's contract allowed her to create new recordings of her entire catalog five years after the original albums were released.
Before Taylor, no one had attempted to exploit that loophole at scale. Perhaps that's because no one had achieved much success with re-releasing songs that everyone already knew. But through a combination of adding new songs ("From the Vault") and appealing to her loyal fanbase, Taylor made it work better than most people imagined possible.
She has since re-released four of her first six albums as Taylor's Version (two of them coming in 2023), with two more likely arriving soon. All four have debuted at #1. The most recent and most successful was 1989 (Taylor's Version), which sold 1.653 million units in its first week. That's larger than any other album since 2015, and significantly larger than the original release, which opened at 1.287 million units.
So it remains to be seen whether Taylor has created new opportunities for artists to take more ownership of their work, or whether she's only exploited a temporary loophole for her own advantage. But either way, she's definitely disrupted the industry.
At a time when the news is full of doom and gloom, and the country is more polarized than ever, Taylor is the one major person who is nearly impossible to hate. And that makes her the Person of the Year that we need right now.
Better in what way? If Time says the most important person this year is a pop star who has done nothing but be really good at pop starring, that shifts their brand towards Teen Vogue celebrity gossip. If they name the head of an $86 billion tech startup promising to change the world, that's a very different brand.
I have no idea which is the better business strategy for Time, maybe the celebrity gossip market is hot right now, but I'm pretty sure the magazine's current brand is a lot closer to Altman than Swift.
But Time's Person of the Year hasn't been serious in a long time. In 2001, they made it Giuliani rather than Osama Bin Laden. In 2003 it was "The American Soldier," in 2005 it was "You."
In 1988 they gave it to the planet Earth considered as a whole. Raising the question of why, if the entire world can collectively count as a single person for the purposes of the award, it doesn't win *every* year.
Since you are clearly a big fan, and I barely know her music from snippets on the radio, you seem like the guy to ask this, because I'd have no way to tell. How well did the re-recordings match the originals? If you spun her records over and over for a summer of your youth, I think you would notice any deviations from the original. Even the digitally-created audio would sound a little different unless she just outright copied the original files and used the exact same software. Anything recorded on real instruments would be even more impossible to match, with the absolute best session musicians playing the exact same instruments and using the exact same sound equipment you could get close but it would be noticeable.
I bought an iTunes version of Belle and Sebastian's classic LP "If You're Feeling Sinister" a few years ago and was bothered to distraction by the very slightly different mixing of the horn parts in "Stars of Track and Field", so maybe I'm an outlier, but if I've heard a song a lot it's like the uncanny valley effect to hear a mix that is only slightly different. I had to go back and buy the original CD off eBay, after checking catalog numbers to make sure it was the original pressing.
The re-recordings are intentionally very close to the originals, but they aren't identical. I would say that the main differences are that Swift's voice is more mature (particularly in comparison to the early albums) and the mixing is slightly better. There are a few tracks with very obvious changes, e.g. "Girl at Home" has different instrumentation; the lyric "She's better known for the things she does on the mattress" is cut from "Better than Revenge".
Taylor's music is deliberately crafted to be easy to sing along with, and that would be jarring if there were many changes. So most of the time, she stuck to the originals as closely as possible. In fact, for me it's sometimes hard to notice the differences.
But then I only started getting into Taylor in 2015, so perhaps someone who grew up with her music would think differently.
I'm still confused by your last paragraph, though. What is the internal experience of being a Taylor Swift fan? I'm sure she's a very good singer and songwriter, but are people attached to her? Isn't her personality "mean girl who snipes at her ex-boyfriends"? Is this psychologically compelling enough to create an empire?
This is like the 5th Swiftie testament I've come across, and they all sound like an infomercial trying to sell me Taylor Swift in 7 easy instalments, or a letter asking for Taylor Swifts being Taylor Swift grant to be extended .
And the other things I have seen them doing: Get products offensive to their sensibility cancelled in an especially obnoxious way (dude, there's like 30 million of us. Do you have any idea what we could do without even trying hard? Take it down.), and wierd uncanny stuff like blank out on her concerts, and just post literal cheerleader chants when her name is mentioned.
I find her face and her name ugly and graceless. She released an anti-patriarchy song lamenting all the possibilities closed to her for being a woman. The messiah thing she must be cultivating on some level, and I don't like that either.
I hate how she doesn't even look any different to when she was first famous, at least Madonna pinned down distinct time eras every time she rose, Taylor Swift's annus mirabilis is just the last word in our culture being irrevocably stuck. Given her outsize influence, I'm going to unfairly blame her directly for this. This is what she wants. It's what she's wanted all along.
"At a time when the news is full of doom and gloom, and the country is more polarized than ever, Taylor is the one major person who is nearly impossible to hate. And that makes her the Person of the Year that we need right now."
Me. I have decided I hate Taylor Swift. What's more I hate every single thing I know about her. So I will hate her enough for everyone.
I was writing a case for Taylor Swift as Person of the Year. I think she deserved it, but from what I saw here and on Manifold it seemed like most people didn't understand why she was even in the running. I wasn't trying to write an unbiased review.
In fact, we might actually agree on more than you think.
- I agree that her fandom is pretty toxic at times. That's not Taylor's fault, per se, but she sometimes takes advantage of it. For example, I felt like she should have told her fans to stop sending death threats to Scooter Braun. But it gave her leverage, so she said nothing.
- I assume by her "anti-patriarchy song" you're referring to The Man. I think she's probably right that she is scrutinized in ways that male singers generally aren't. But I find it a little cringy how hard she leans into the idea that being a woman is a disadvantage, while at the same time shrewdly using her identity as a woman to further her career.
- And yes, being famous is exactly what she wanted all along. In some of her songs, she's very self-aware about this. For example, take this line from Mastermind:
"No one wanted to play with me as a little kid
So I've been scheming like a criminal ever since
To make them love me and make it seem effortless."
You have to read between the lines sometimes to tell which of Taylor's songs to take seriously and which ones are a joke (e.g., Blank Space). But I'm confident that this one is absolutely serious - she's obsessed with fame to an unhealthy degree, and she knows it, but she can't stop.
So it's actually fine with me if you hate her. I don't think she's a saint, but I do think you're missing out.
> For example, I felt like she should have told her fans to stop sending death threats to Scooter Braun. But it gave her leverage, so she said nothing.
I am not sure if she counted this as a leverage for present or as a revenge for past (not that either is commendable, but both are somewhat understandable).
Taylor definitely isn't considered a mean girl by her fans. I asked my wife, and she called her a nerd. She was a band geek in high school that no one liked, and now she's an amazingly successful business woman who has upset industries and boosted the economy in presumably big ways.
"I'm sure she's a very good singer and songwriter"
This is a bit of a tangent, but I think to understand Swift's success, you need to appreciate her showmanship. I would say the order is songwriter > showman > singer. This leads to some confused people asking why she's so successful when her singing is no better than many others.
If you want to understand her success, you really ought to see her in concert. There are many Swifties, and no doubt they have a wide range of experiences, but I would say an important element for many is having shared in this extraordinary experience. Swift has a remarkable ability to create a feeling of intimacy with 50,000 people.
Unfortunately, tickets to her concerts are notoriously difficult to obtain. Fortunately, it's her birthday in a week, and the Eras film will be streaming on various platforms (e.g. Apple, Amazon, YouTube). I should warn you though: it's very long, even longer than the theatrical version.
That's a great question. I can see how you would get "mean girl who snipes at her ex-boyfriends" from certain media portrayals, but she's generally not mean-spirited.
I think what makes being a Swiftie unique is that Taylor is in love with her fans at least as much as they are with her. In fact, I suspect that's the reason many of her actual relationships haven't worked out - no guy will ever be first in her life, because her fans will always be more important.
For example, she's always made herself as accessible as possible. Her mom used to locate her superfans at her concerts and invites them to come meet her (though I'm not sure if that still happened with the Eras Tour).
After 1989 was released, she made this video of sending Christmas presents to her fans: https://www.youtube.com/watch?v=j3yyF31jbKo. And they weren't generic gifts - she stalked all of them on social media long enough to pick out a unique and special gift for each person.
And she constantly shares intimate details about her relationships through coded messages in her album notes. Taylor wants all her fans to know who each song is really about, even if she doesn't directly name the guy in the lyrics.
With that context, it's easy to understand why she and Joe Alwyn broke up:
One of the best quotes from their anonymous source is, "[Taylor and Joe] had plenty in common and fell in love in a safe bubble while she was retreating from the world during Reputation... Then the pandemic hit, and they were locked down together and able to continue growing their relationship in this insulated way. But he didn't really 'know' her yet outside of that bubble."
And also: "Joe has struggled with Taylor's level of fame and the attention from the public."
That's obviously the way Taylor wanted people to see things. But reading between the lines, I think the real issue was that he wanted to remain first in her life, and she cared more about preparing for the Eras Tour. (Notice how she never seemed that disappointed by their breakup - she got what she wanted.)
So to sum up, the internal experience of being a Taylor Swift superfan is that the most powerful and most socially intelligent person in the world loves you deeply, and is constantly trying her best to show it. It's not so far off from the way Christians see Jesus.
And to be clear, I don't count myself as a superfan. I'm a "huge Swiftie" compared to most people here because I can sing all the words to most of her songs, but I wouldn't call myself that in a group of Swifties, because I don't worship her the way many of them do.
Elvis was before my time, but it could be similar.
Maybe in 50 years people will go to Vegas to be married by Taylor Swift impersonators singing Love Story. That sounds way better than being married by Elvis. 😂
One more item about how/why Taylor may not come across as a 'mean girl' is that she shows some awareness that the problems aren't all her boyfriends fault.
I agree with you, but I think Blank Space is a poor example. Taylor has said that it's a parody of how she saw herself being portrayed in the media - it's not an actual confession.
The best case for Taylor as times person of the year is that it gets the Time's magazine the most attention.
Taylor is the anti-mean girls. The "you will always have friend and community while maintaining your unique specialness" product. They maintain an image, communicate the messaging people want to hear, and sell the product. Their songs are intentionally lower skill for singing along to take this even further. She occupies a niche and is not a reproducable approach to succeeding in the industry.
I don't think the Times is dumb enough to actually think Taylor is the person of the year for her own financial success (mega tour wow, omg congrats you made a movie too, AND you big dicked the big dicks!?!?). Times is smart enough to know that the swifities will bandwagon the decision and give Times tons of free media. Other people will question it with good reason and that will incite drama, more convo, more free publicity.
I wish I could send free psychadellics to every swiftie in america
> Why is Taylor Swift so high? I understand she’s a very famous pop star, but hasn’t she been an equally famous pop star every one of the past ten years?
I don't know why, but anecdotally, I've been hearing way more about Taylor Swift in the past 6 months than I did in several years before that. Seems she had some sort of resurgence in popularity.
Regarding long-term markets, I had a thought on early resolutions. Specifically, I created a market that asks if a large construction project will finish in time. It will resolve as NO as soon as a delay is announced. This is good because it reduces the expected period of capital lock-up. This is also bad because it creates different incentives for different participants: If you expect major delays, you also expect a shorter lock-up. If you expect a punctual finish, you expect a longer lock-up.
I'm basically resolving the market early based on a (somewhat authoritative) third-party estimate. This is problematic in itself, but seems fine to me because the project management organisation is strongly incentivised to not falsely announce delays.
I was wondering if anyone had thoughts on this approach.
"The best realistic medium-term outcome I can imagine for the people of Gaza is as something like a West Bank without settlements and roadblocks"
I don't think this is even remotely realistic. Israel has worked_ hard_ to cleanse the good areas for colonization and settlements - there is absolutely no way they would ever return all of that, not even if a credible peace was offered (and no-one can offer that). That ship sailed long ago - it could _possibly_ have happened in the early 2000s, but not now.
Thanks - re-reading I _think_ what Scott is saying isn't fixing the West Bank, but having Gaza in the same kind of statelet. That's probably doable, at least after Israel strips away a chunk along the border to get a wider (and obviously necessary) security zone.
So mostly status quo ante bellum (apart from, ideally, no Hamas for a while), alongside _some_ territorial losses. The blockade will obviously remain, as will even more draconian entry checks into Israel.
I don't know much about the Middle East - why will the blockade obviously remain? I thought the blockade was in place because Hamas kept converting imports into rockets. If Israel has enough control to keep the terrorism down, couldn't they relax the blockade?
Israel _won't_ be able to control terrorism. They might (possibly) be able to ensure the terrorists aren't actually in charge, but terrorist groups are going to remain and pop up.
Two possible thoughts, one using technology that I'm not sure of:
1) If lie detectors work today, have each Gaza or West Bank resident swear that they will never attempt to kill Israelis while on a lie detector, and allow the peaceful ones to stay.
Alternatively:
2) Have a two-state solution, but the Palestinian state is not self-governing, but is rather administered by the People's Republic of China, analogously to Xinjiang. Admittedly the Uighurs only killed ~100 Han Chinese in their attacks, a dozen times less severe than 10/7/2023, but the PRC seems to have the will and the technology to manage Islamic minorities safely at this point. I think the PRC could prevent a Hamas2.
Lie detectors detect lies more accurately than chance (which is better than most humans), but are mostly useful as security theater. They're not nearly accurate enough to do 1 particularly well.
(1) I am totally uninterested in talking with, using, or playing with ChatGPT or any of the competing chatbots. "But you can do art and write essays and this, that and the other!"
Yeah. I find I don't care. I tried one model that allows you to just start online, and I found myself sitting there with a blank input prompt box, because there was nothing I wanted to ask it, no conversation I wanted to begin, not even any "can I make it do no-no things?" because I don't care what a machine spits out as 'Hi I'm your real liddle fwend who wanna talk wid u about stuff just like human person!'
So I'm approaching this whole giant blob from the point of view of "uninterested idiot"
(2) Given all the chit-chat about OpenAI, here's what Microsoft are doing in the field. This isn't theory, this is "selling a product to businesses". I think it might be interesting to get the view from the outside, as to what the actual implementation of AGI is visualised as (not the theoretical paperclipper/fairy godmother stuff):
That mess above brings you to the PowerPoint Presentation from the November "Microsoft Discovery Day: Building AI-Powered Apps". It amuses me, a bear of very little brain, that nestled in among all their salesmanship about how ground-breaking the AI is, there is still an AI-generated piece of art which can't understand how to spell correctly "integration" or "recognition" in "AI integration" and "Voice recognition" text on image.
"Microsoft and OpenAI partnership
OpenAI: Ensure that artificial general intelligence (AGI) benefits humanity
Microsoft: Empower every person and organization on the planet to achieve more
Azure OpenAI Service :
GPT-4 & GPT-3.5-Turbo Text, Chat & Code
DALL·E 2 Images
Whisper Transcription & Translation"
Very nice, how do they see this translating into practice?
"Introduction to Top Use Cases
(1) Business Problem
Productivity is lagging
Business Needs
Increase Productivity
Solutions
• Conversational Search/Knowledge Insights
• Code Generation and Documentation
• Trend Forecasting
• Report Summarization & Generation
(2) Business Problem
Need for process Automation
Business Needs
Automate Processes
Solutions
• Document Processing
• Workflow Management
• Fraud Detection
• Supply Chain Optimization
(3) Business Problem
Degraded Customer Experience
Business Needs
Improve Customer Experience
Solutions
• Intelligent Contact Center
• Agent/Employee Assistance
• Virtual Assistance
• Call Analytics
• Call Summarization
(4) Business Problem
Creating Content is Time Consuming
Business Needs
Build Creative Content
Solutions
• Marketing/Sales Content Generation
• Personalized Content Generation
• Product Design & Development
• Digital Art
What can Generative AI Do?
Generate New Revenue Streams
Deliver Differentiated Customer Experiences
Modernize Internal Processes"
In future, Scott can be freed from the time-consuming burden of "build creative content" by outsourcing it all to AI. How nice to look forward to this!
So the AI future in the workplace is not going to be increased leisure and money, it will be "you WILL be more productive! you WILL increase velocity! if you still have a job, that is, and have not been replaced by the machine altogether". I honestly think *that*, rather than "unaligned AI will find a way to 3-D print neurotoxins to be delivered globally by nanobots to wipe out humanity" is the real problem we will be facing and should be discussing.
"The only error is that it recommended a deli counter as if it were a sit-down restaurant."
Be the change you want to see in the world! Enable sit-down dining at the deli counter now! Grab a few rolls from the bakery section, a bottle of the beverage of your choice from the wines and spirits, and there you go!
I can't understand you not wanting/needing to experiment with it. As far as not knowing what questions to ask, you can ask it the same things you would ask Google, or any other search engine. It isn't about conversation, but about finding answers and doing language busy-work.
Of course, it is about as accurate as any source found on the internet, and you must bear that in mind with any output.
But it can produce average quality written material, without typos, or standard messages. Here are some things you can ask it with good results (after critically reviewing and revising the output):
- Give me a job description for a bookkeeper position that does both accounts payable and accounts receivable work, among other things.
- Why does ice take up more volume than water?
- Write a two-page essay about why dumb phones can be better than smart phones.
I think, looking at my own reaction, it's because I don't see it as anything other than a machine.
So I'm not interested in whatever it outputs, since I know it's been formatted to pretend to be a 'personality' and reply to me that way.
If I want to look something up, I'll go straight to Google, and there isn't any "language busy-work" I need it to do; I'm long past having to write essays for homework, thank God, my workplace has standard job descriptions in place if it is hiring on new staff, and for work it's generally short emails in response to boss or colleagues.
Now if I needed to churn out blah blah meaningless reports (like a previous job) I might be happy to hand that over to it, but right now I don't, so there's nothing there I care about using.
I'm very surprised by this myself, since I would have imagined I'd *want* to experiment with it, but I think part of the hype has put me off and the rest is "yeah - it's a box, it's not thinking or understanding, why do I care? I don't".
Of course it's a machine. Those who think it can actually think or has consciousness I believe are misguided. But you probably have other things you may want help with, or which could spur your own creativity.
What about writing a thank-you note to someone you don't really know? What is an appropriate congratulations message to an acquaintance? I'm sure you get stumped on various kinds of writing like everyone else does, but the machine is never stumped.
"What about writing a thank-you note to someone you don't really know? What is an appropriate congratulations message to an acquaintance?"
"Thank-you notes to people I don't know" are part of the bread-and-butter of my current job, and I have it reduced down to a formula which I think nicely balances oily flattery with not seeming too obvious about it 😁 By the time I've prompted the ChatBot, removed the Americanisms, and read over and corrected the thing, I'd have the letter written.
Too much information about my personal life, but I don't write letters, notes, emails, etc. to acquaintances and friends 'cos I don't got none. All that writing is on the job, any 'personal' stuff is via social media and I curate that kind of thing I say along strict parameters. EDIT: There are also plenty of old-style letter-writing guides and new model letter examples online, so while I could find one of these AIs to do it for me, it's just too much bother. "Hey, Chatty, here's the person, here's the sitch, here's what I want you to say, make it informal and breezy, yadda yadda yadda", get the output, keep going through it, finally get a finished product to bang off - I could, but that sounds like climbing onto a pair of stilts to walk up a ladder in order to reach the top shelf of the kitchen cupboard, you know? Quicker and easier for me to go on my tippy-toes.
When I'm writing for myself, as in fiction writing, I prefer to do it my own way. Having the bot churn out some pap that's tailored to "and you said you liked marmalade, rain on the sea, and violent torture of the woobie" to 'personalise' it to my requirements - no, thank you!
You can’t think of *any* busywork you’d ever want to outsource? LLMs are the written-language equivalent of a graphing calculator for math; they are good at doing boring repetitive stuff to save your time to work on the important items.
Even if you don’t have anything like that, surely you can understand why other people would find it useful. For example, I have no interest whatsoever in writing tedious marketing emails, but that could be a big part of my work that I can just make an AI do for me!
Oh, I can see the potential for automation, but right now what it can do is not what I need done, or I don't have the resources to link up an LLM to the things it needs to scan in hand-written documents, etc. and turn those into intelligible data.
The impression I get, and I am probably wrong, is that I have to babysit the thing while it churns out response to the prompt, so it doesn't save me time to get on with doing other things. As it stands, I certainly would *not* trust "get LLM to write that reply email and send it off to Revenue or the Department of Education" without double-checking it first.
> I think, looking at my own reaction, it's because I don't see it as anything other than a machine.
That's not the real problem; the real problem is that it's a *really stupid* machine. ChatGPT is miles ahead of ELIZA in terms of grammar and syntax, but not very far ahead at all in terms of actually delivering useful information. As you'd said, Google can deliver the same data, in a more reliable format.
That said though, I think that image generators *do* have a useful purpose: in our RPG gaming group, we use them regularly to come up with character pogs. Would we still use them if at least some of us had artistic talent ? Probably not, but none of us have such talents, so...
ChatGPT has occasionally done better than Googling stuff myself, because when you're trying to learn something new it can be useful to aggregate the result of many Google searches into a single summary.
For instance, I was looking for a knife sharpener recently, and Googling it turned up a ridiculous number of possible options - do you use electric or manual, ceramic or tungsten carbide or diamond, the kind with spinning wheels or the kind with v-shaped grooves, how much material does it take off, what angle does it sharpen at, do I listen to the people on Reddit saying I need to learn to use a whetstone, etc. So after Googling around for a while, I went to ChatGPT and said "Tell me about the different types of knife sharpener and what the pros and cons are" and got a nice summary of everything.
IMHO, ChatGPT exists to summarize the internet for people, in essay form, and perhaps 99% of the internet is digitized mass communication. It's a real time, automated version of Wikipedia, without the source notes (well, without reliable ones). The way to make the most productive use of it is to carefully craft the prompts to correspond to sources you have already checked. I have no objective evidence, but I'm convinced that it makes the most money for users by being a marketing tool.
If that sort of thing does not interest one, then the use cases go way down.
"Here are some things you can ask it with good results (after critically reviewing and revising the output)"
That parenthetical is doing a lot of heavy lifting, and in a way that negates a lot of the value of LLMs at least for applications I care about. ChatGPT's answers *cannot* be trusted for anything important. We had the example right here of ChatGPT all but advising SpaceX to do, in the name of increased "safety", something that would turn every Starship into the equivalent of a tactical nuclear weapon with a hair trigger.
If the question is simple, and if I am confident that I know the subject, then I can proofread and revise very quickly. But for simple questions in my area of expertise, I can come up with my own answer very quickly, perhaps not quite as quickly as prompting and proofreading and revising ChatGPT, but pretty close and with less possibility of an embarrassing goof. If the question is not simple, or not in my area of expertise, the proofreading part approaches just researching and solving the problem myself.
I can see ChatGPT and its kin as useful tools for generating pointers to answers, but I've already got search engines and wikis for that.
For generating probably-accurate trivia of no particular importance, sure, that's going to be fast and easy, but why should I care?
" We had the example right here of ChatGPT all but advising SpaceX to do, in the name of increased "safety", something that would turn every Starship into the equivalent of a tactical nuclear weapon with a hair trigger." Yup! That was dramatic.
Yes, I've read of it. I suspect that a large chunk of LLMs' current problems come down to (to anthropomorphize) they being insufficiently self-critical. And one can be wrong in either direction, being too self-critical or insufficiently self-critical. And it isn't so much that Project Orion was a _bad_ idea (the expected specific impulse would have been much better than chemical fuels) as that it got nixed by the limited test ban treaty and, by now, the combination of improved chemical rockets and (for low absolute thrust, high total delta-v) the combination of ion or plasma thrusters plus nuclear reactors made Orion moot.
The people proposing Project Orion were not factually wrong about what would happen if an Orion were built and launched. They recognized both the costs and the benefits, and considered the benefits worth the costs. You may disagree, but that doesn't make them or their ideas dumb.
ChatGPT, is just plain dumb. It knows a lot, but it's too dumb to realize that it doesn't know everything, and it has great difficulty learning new things.
But did they really consider the fallout from the explosions? The trail of it through space? Do they also explode nuclear bombs to slow down? Maybe it could be considered as a plan of last resort to escape forever from a doomed planet, but it's hard to imagine actually using explosion after nuclear explosion to power a starship.
I'm mostly probing to see when the LLMs stop hallucinating.
I just tried ChatGPT again, this time asking for the elements, sorted in order of increasing density, and it didn't sort them correctly. (Ok, it was the free version, still, this isn't new or difficult information)
Dont ask it for data. Ask it to summarize data you give it (you point it to where you want it to go--in your case, something like summarize this paper (contains the table of periodic elements)).
Well, I was hoping for better from even "a blurry jpeg of the web". A good chunk of the potential value of an LLM trained on a vast quantity of information is to have it pick out data from a hundred places _without_ the user having to explicitly specify those places. If it can't do that, so be it, but then the technology is significantly less valuable :-(
The problem here is that the vast quantity of information that the LLM is trained on, the internet, does not itself consist of reliable answers. In fact a fair chunk of it is misinformation. So long as we are the ones training it, it isn't ever going to be more accurate than we are.
More precisely, where there is general agreement by writers of the training set on some question, if the agreed answer is wrong, the LLM will get it wrong as well.
As LLMs that I've been able to play with recently stand, the LLMs behave substantially worse than that. In the example that started this subthread, I asked an LLM (Bard, in this case) to list the elements in order of their densities (of their condensed matter phases). This isn't the sort of data where there is a lot of misinformation on the web. The right answers are pretty much in Wikipedia, and, as far as I know, there aren't widely promoted wrong answers. In fact, the _numbers_ that Bard came back with looked about right. Weirdly enough, it got the _sorting_ wrong. So this is a different problem from misinformation on the internet.
In principle, an AI that uses the internet as a training set can do better than the raw error rate of information on the internet by looking for inconsistencies. If it sees five people make one deduction from a set of premises, but they all have logical errors in their arguments while a sixth person starts from the same premises but makes logically valid deductions and arrives at a different conclusion, a sufficiently sophisticated training process could theoretically notice this and arrive at better conclusions than raw statistics would yield.
I'm still waiting for worse blunders to be fixed. :-(
(1) For me it is simple; ChatGPT replaced a part of search engine usage. If I am looking for a specific web *page* (for example a YouTube video), or I want to read a *document* in its entirety, I use Google. If I am looking for an *answer* to a question, I ask ChatGPT instead.
Ultimately, using a search engine to find answers is very ineffective, because a search engine only finds *keywords*, and those can appear in unrelated context, and can be gamed. In best case, you still have to click on the page, click through the pseudo-GDPR "consent" forms, scroll down, and find the answer in text. ChatGPT instead answers your question directly, immediately, using the knowledge from many web pages; sometimes it even provides additional useful information.
It does not have to be super-intelligent to be smarter than Google. Some things are so SEO-spammed that it is virtually impossible to find the answer by googling. Try e.g. "Japanese words starting with v"; if you find an actual answer on the first page of Google results, I am jealous. ChatGPT just gives you a list.
For a software developer, ChatGPT can also done some "manual" work for you. Something that a junior developer could get done in an hour, except that ChatGPT does it in a second.
(2) Sure, "2x productivity" has always meant that you will have to produce 2x more in a unit of time. Not that you will have 2x higher salary, nor that you could get a part-time job now. For an employee, the only difference is having to learn how to use a new tool (preferably before the job interview, during your free time).
The problem with ChatGPT as a replacement for search is that sometimes it thinks it knows the answer but it's confidently wrong. So if it's important you have to look it up anyway. If you use the web-enabled version it's even worse because it just gives you a summary of the top few bing results even if they suck. I agree that it can be good for certain questions that have no good search keyword, or where no high ranked page has an actual answer.
I think ChatGPT (particularly the speech version) is useful for satisfying casual curiosity. If I'm reading about, say, the constitutional referendum in Chad, and I'm not familiar with the background, I can just ask it, get an answer, and continue reading.
Did you know that purple used to use straw, also called "thresh", instead of carpet? It was spread in a thin layer on the floor. To keep it from spilling outside, they put a bar at the bottom of the door. This was the "thresh-hold".
Except that isn't actually true. But I bet it's easy to find that story on the internet, in lots of places. ChatGPT might even give you that story.
Don't believe what you read on the internet until you can confirm it. ChatGPT isn't more trustworthy just because it's a machine.
"Sure, "2x productivity" has always meant that you will have to produce 2x more in a unit of time."
Which is how new tech has always been implemented, from the invention of the typewriter onwards. But it gets *sold* to us, the cogs in the machines of commerce, as "this will lighten your workload and reduce your work day". Nope. Free time remaining now the new wotsit has sped up production and whizzed through what used to take four hours to complete? Don't imagine you can clock off early, now you need to do 2x the work in the same time.
> Don't imagine you can clock off early, now you need to do 2x the work in the same time.
This is a logical consequence of the global elite's century+ run of spectacular success at making sure that no appreciable increase of abundance ever happens to the actual necessities of life for which "rat race" inmates (read: you, I, and everyone you're ever likely to meet) "race to the bottom" in competition: comfortable housing; freedom from criminal predation; such food as existed 300y ago (i.e. not Velveeta, "McFood", antibiotic-laden meat, palm oil "chocolate", etc) ; health (which is not 100% equivalent to "access to medical treatment", as Americans were led to think) ; insulation from financial volatility (i.e. "sorry, you don't get to have a roof over your head now because someone-on-the-other-coast-did-something-with-stocks" etc.)
See also Moly's "involution" :
"Involution is basically…a race to the bottom, stressful competition with no winner, basically. The best explanation I’ve seen of it is that if everyone was sitting in a theatre watching a movie, and all of a sudden, to get a better view, somebody stands up to watch. So the guy behind him has to stand up, and the guy behind him has to stand up. And soon enough, everyone in the theatre is standing while watching the movie. Everyone is less comfortable, they’re being made to work harder, and yet the movie is still the same movie. They didn’t get anything extra out of it at all." ( https://weibo.substack.com/p/110323-faq-answered )
... or Scott's "Moloch" piece.
"Involution"/"Moloch" is not in fact an inescapable natural phenomenon. It is engineered -- by the people who prevent the proverbial theater audience members from whacking the first stander-up on the head with a heavy blunt object. Which people? Our rulers. In every case.
But most pages for "Japanese words starting with v" -- heck, most pages with TITLE "Japanese words starting with v" -- actually contain *English* words starting with v. There are tons of such pages, probably for Scrabble players or something like that.
ChatGPT explained that only the loanwords starts with v, and then gave me some examples.
I'll reiterate a "prediction" I've been offering recently: We'll arrive at a substantial understanding of the internal structure and operations of LLMs BEFORE (the mythical) AGI is attained. Moreover, once that happens, the need for belief in AGI will weaken and the idea will disappear, as did the idea of phlogiston once upon a time. What will replace AGI? I surely don't know, specifically. But in general, a richer and most robust understand of the mechanisms of both artificial and natural intelligence.
The concept of "combustion" didn't vanish from our vocabulary when we discovered that phlogiston wasn't a real thing, because "things burn" is still a real thing and "combustion" describes that, regardless of if the mechanism is phlogiston or chemical energy.
Similarly, even if it turns out that we're wrong about the nature of consciousness and how LLMs "think," the phrase "general intelligence" isn't going to vanish from our vocabulary, because "the sort of thinking that humans can do and current-generation LLMs can't" is a real thing, and you need a word for that no matter what the underlying cause is. It describes a capability, not a mechanism.
What's going to happen is that we're going to come up with mechanisms to account for capabilities. In the process, we may have to revise our sense of what capabilities are possible. Just because we have a name for a capability doesn't mean that it's a real thing. If we don't know how it works, then there's a sense in which we don't know what it is.
Yan LeCun keeps making the point that, no, we don't have general intelligence. We can do a lot of things, but not everything. Moreover, each individual has specific capabilities.
So, what cognitive capabilities are humans lacking that we can't be called "general intelligence"? Because it sure seems like our intelligence can be applied to a broad variety of fields. We don't have a specific lobe of our brain that enabled us to invent quantum physics, we have a general process of understanding the world which works for every field of science.
If an AI is able to do everything that an average 100 IQ human can do, without needing to be retrained between tasks, I think most people would call that "general intelligence," regardless of if Yan LeCun thinks there are still secret cognitive tasks to discover which neither humans nor machines can do.
Yup. I view "able to do everything that an average 100 IQ human can do" as a goalpost (albeit one with some haziness). How much detailed understanding of mechanisms (beyond training neural nets) is needed to reach that goal is an open question.
Where's the 'intelligence' reside? In the 'raw' human brain, or in the brain that's been specialized of this or that subject? Is your brain specialized for every subject under the sun? Is anyone's? What are you imagining for the AI of the future? One giant humongous 'intelligence' of 10,000 AIs communicating with one another, each of them variously specialized?
This isn't a serious intellectual discussion. It's bar-room chitchat. That's fine in its way, but...
Just because we can string a bunch of words together in a coherent way doesn't mean that they refer to something that's real, if not in the past or present, then possibly the future. LLMs do it all the time, but sometimes they're "hallucinating," "confabulating" is a better term, BTW. And that's what talk of AGI etc. is, confabulation, a collective confabulation.
Your prediction isn't relevant unless LLMs scale all the way to AGI/ASI.
So with that in mind I predict that your prediction will be wrong in precisely the scenario where it matters. Pretty decent chance of you being correct in the case where it turns out to be irrelevant though.
Do you think the successful AGI/ASI scenario will happen With or Without machine learning? We aren't going to get there unless we understand what's going on inside ML models. Basically, we're dealing with associative memory. Details will vary, etc. But the principles are the say. We just have to figure out what they are.
I revise my prediction to a general understanding of the mechanisms of ML models must come before AGI/ASI.
I don't know if you are right or wrong, but I can think of an alternative: We get AGI when we figure out what conceptual self-awareness really is.
InB4 someone says SA is simple and we already understand it: obligatory reference to the XKCD cartoon of the physicist wondering why other fields need whole journals for them.
"Just model it as <Simple Object>, and then add some secondary terms to account for <Complications I just thought of>. Easy, right?
Right, just what does that mean ("AI that seems about as smart as a human")? How are you going to specify it? An IQ range? Scores on college entrance exams? Maybe you just talk with it and arrive at an intuitive judgement.
Refresh yourself on the initial remarks Steven Pinker made in his debate with Scott Aaronson in 2022: https://scottaaronson.blog/?p=6524
Well, at the very least (and I am very much *Not* proposing that this is all we would need to do, just this as a first tentative step) we would need to create the functional equivalent of the networks of associated schema and prototypes that the Long Term Memory uses to store the idea of every life experience we have had that meant something to us.
To set the record straight: I once gave a ~2% probability for the classic AGI-doom paperclip-maximizer-like scenario. I have a much higher probability for an existential catastrophe in which AI is causally involved in one way or another — there are many possible existential catastrophes (nuclear war, pandemics, runaway climate change…), and many bad people who would cause or fail to prevent them, and I expect AI will soon be involved in just about everything people do! But making a firm prediction would require hashing out what it means for AI to play a “critical causal role” in the catastrophe — for example, did Facebook play a “critical causal role” in Trump’s victory in 2016? I’d say it’s still not obvious, but in any case, Facebook was far from the only factor.
I would argue that there is pretty solid journalistic level evidence that it was a contributing factor (and not by accident either--it happened on purpose). Given how close the election was...
Also, the GREENS took some votes from Hillary Clinton. And guess who had lunch with their candidate? Vladimir Putin. Misha Frier: "Jill Stein launched the 2024 White House bid as a Green Party candidate. Is she getting any support from Mother Russia again?
Back in 2015, after the Annexation of Crimea and the launch of the war in Donbas, Jill Stein was at a gala dinner in Moscow to celebrate the 10th birthday of the Russian TV propaganda network Russia Today. She was there at the invitation of Vladimir Putin and sat behind the same table with him and two former KGB agents, Gromov (chief of propaganda) and Ivanov (Putin's chief of staff), at the gala dinner.
In 2016, Stein's presidential campaign was heavily promoted by RT. I mean what better person to ask for endorsement of an environmentalist candidate that a dictator who bases his power on oil and gas sales to the West?
To be clear, your 2% risk is from being paperclipped (inner alignment failure + fast local takeoff?) and the probability above 2% comes from exacerbating non-AGI risks? Does that mean you don't have much room for non-paperclip AI catastrophe, like rapid value drift from substituting ourselves out of the economy or fast local takeoff from a badly outer-aligned AGI, etc?
> I thought Kalshi was trying to cultivate good will with the CFTC, and this seems pretty adversarial.
A couple of perspectives. First, cultivating goodwill can be (is?) adversarial. If nothing else, you're competing (sometimes with the CFTC) for the CFTC's attention.
Second, even if the parties are getting along, an adversarial process might be inevitable to break a logjam. When an agency reaches the limit of its statutory authority, its hands might be tied (especially if there's political or media scrutiny). In that case, you go to Congress for a new statute or you go to the courts for a new understanding of the statute. Either way, adversarial.
To be clear: I'm not an expert in CFTC, prediction markets, or regulatory law in general. But I am a lawyer at a state-gov agency. We work very closely with regulated entities and have a strong customer-service culture, but it's often clear that an entity has to turn to politicians or judges to get around limits on agency power.
I've seen that from the other side in a previous job; the regulations based on the relevant legislation can be interpreted in two ways; the wording is vague enough that either interpretation is plausible; and nobody wants to be the one to stick their head on the chopping block* and say "yes, make decisions based on interpretation 1/interpretation 2", all the way back to the head office in the government department, until there's been a law suit that results in a ruling that "interpretation X is the correct one". After that, you can't get into trouble, because "well the courts told us we had to do it this way".
*Because if you say "do it by interpretation 1", you are spending more money, and the Department will not be happy with you and you will get yelled at, you will get your bosses yelled at, and your bosses will yell at you about over-spending the budget
If you say "do it by interpretation 2", you don't spend the money but then the aggrieved client who was refused may well go to court to force the local authority to spend this money. This is good in that it does establish a precedent, but bad in that you will get yelled at and get your bosses yelled at for dragging the Department into court.
"California City is a city located in northern Antelope Valley in Kern County, California, United States. It is 100 miles (160 km) north of the city of Los Angeles, and the population was 14,973 at the 2020 census. Covering 203.63 square miles (527.4 km2), California City has the third-largest land area of any city in the state of California, and is the largest city by land area in California that is not a county seat.
Much of the workforce of Edwards Air Force Base, which is located 18 miles (29 km) southeast of the city, is made up of city residents. Other major sources of employment include California City Correctional Center (California Department of Corrections and Rehabilitation); Mojave Air and Space Port and its flight test operations; and the Hyundai/Kia Proving Grounds located in the rural southwestern part of the city. California City has a park, a PGA golf course, and a municipal airport."
Now I want to know, is there a California City *not* in California?
I'm entertaining the idea of trying manifold.love, but I have a physical preference I perceive is the opposite of what people would expect. It seems like that would be important information for anyone to make better bets, but I normally avoid putting that publicly on my profile in dating apps because because I'm *open* to women who don't match that preference and I don't want to burn that bridge with them (for them to know I have a preference they don't meet). How would it make sense to handle that on manifold.love? Maybe accept that this is the one dating app where I should risk ruling out all women in one category for the sake of better matches?
I think the details of Kalshi's proposal for election markets are very important. iirc, the original proposal was markets on the outcome of Congressional control with position limits of $25,000 -- seemed fair enough. But their most recent proposal allows ECPs (eligible contract participants) to invest up to $100 million, and I think this is just way too high a limit to start playing with, and I worry concerns about manipulation actually become quite plausible. I get why Kalshi did this, because the CFTC said that they didn't believe the contracts had significant economic value for hedging with a 25K limit (which oil company is going to stake 25K on an election market to hedge against Democratic control), but I think the better route is to start small, and see what problems arise, instead of just opening up election betting on such a massive scale all at once. I reckon it will probably not end up being a good idea to have such high position limits to start, if ever.
If the court rules on the narrow question of whether "election markets" are or are not "gaming", that could still give the CFTC enough leeway to regulate contracts for their manipulability (warning: not at all a legal expert). But if the court just totally guts the CFTC's ability to regulate election markets, I think this will be bad -- I think if you want markets to work well (not just in terms of accuracy but social legitimacy), they need to be regulated (imposing requirements like KYC, market surveillance, etc.).
"But their most recent proposal allows ECPs (eligible contract participants) to invest up to $100 million, and I think this is just way too high a limit to start playing with, and I worry concerns about manipulation actually become quite plausible."
Well how can that possibly be risky, it sure isn't like we've seen a magic money machine founded on Rationalist pet project lines with big sums of money floating around going very publicly "ka-boom!" in the press recently with attendant trials upon trials 🙄
Scott: "An Israeli occupation would involve constant bloody resistance; I'm not convinced it would be any better for people on the ground than Hamas (though someone can try to convince me otherwise)"
Well, one possibly convincing argument is that we have a test case for what it means for Israel to occupy a chunk of Palestine with a couple million Palestinians in the 21st century, and it sure seems to be better than Hamas. Israel has occupied the West Bank continually for decades, with soldiers, police, and settlers. There has been resistance, and I suppose pedantically you could call it both continuous and bloody, but I'm guessing you were thinking of something more than that.
And while I suspect that both Israel and the non-Hamas Gazans would prefer that Gaza be occupied by UN peacekeepers or maybe even Egypt, but those aren't likely to happen. But I don't see any fundamental reason why an Israeli occupation of Gaza, particularly in the medium term, should be any more bloody than the West Bank is now.
On the one hand, at least in the short term the Israelis have given the Gazans cause to be really, really upset with them. On the other hand, the Gaza population is presently going through a filter that will select for not being eager to engage in bloody violent resistance to Israel. And there are no settlers in Gaza, nor are there likely to be, which eliminates a huge ongoing aggravation.
I just fixed the issue with the embeds. There was a difference in the embed graph code that was making it connect the points for each bet with straight lines. Which is wrong! We put every single point* so the chart should be stepped rather than linear interpolated. Usually this wouldn't be a significant visual difference but in this case nobody bet on the market for an entire month before it dropped.
*[skippable nerdy aside]: We used to literally put every point on the chart but since lk-99 this would be really slow on the most popular markets. And on a chart that is at most 700px wide people can't perceive the level of detail from tens of thousands of bets anyways. At the same time I HATE those smooth curvy trailing average charts that ui designers keep using because they don't understand that data should not be beautiful; data is bumpy, sharp, and real. It took some tinkering but the solution I settled on for big markets is to slice the x-axis into 500 evenly-sized buckets; take the min, max, and median in each bucket (or the last value in the previous bucket if this one is empty); and draw the chart with those points. I am pretty proud of the end result, I think it looks very close to what it would look like with full data. As long as you don't zoom in :p
Oh, also we start the calculation with no more than 50,000 bets on a market, semi-randomly sampled, because I was too lazy to paginate through all the bets and eh it works
Manifold.love is a neat idea. I'm all in favor of anything that comes close to bringing back 2005-2010-era OKCupid, and the matching questions are a big part of that.
That said, I'm dubious of the "make a Yes bet to send an intro" mechanism. It suggests some specific target user demographics: people who have more than a 50% chance of a first message turning into a first date. As far as I know, that niche contains most women and very few men.
Are there guys on the platform making confident Yes bets of any substantial size, or is everyone just putting in Ṁ10, or are people spending real money to buy Ṁ to make bigger mostly-losing bets on intros, or... something else?
I don't think a Yes would have been a good bet for any message I've ever sent on a dating platform, or any conversation I've struck up with someone IRL, including people who later became my partners or my wife. I predict this is the case for... almost everyone?
My prediction for Gaza is that no government will reign there. Israel will conduct frequent raids in an on-going basis but based from outside the Strip. Hamas will more or less retain military control of Gaza but by having to operate from underground will not be able to effectively govern the territory. No other force will be willing to operate a government in Gaza. — 75% chance.
The UN and/or Red Crescent will become the de facto government of Gaza. — 20% chance.
Basically there will be no police type activity in Gaza from now on. Just military and basic humanitarian activity.
I'm the #6 most profit on the POTY market, so I think I'm qualified to speak on how I got it right.
The Eras Tour was the highest grossing tour in U.S. history. It generated small booms in excess consumer spending in every city where she performed; According to current estimates, about $4.6 billion in excess consumer spending. It's like Swift ran her own private stimulus program for post-COVID downtowns. If you worked in or near a city she performed in it was all people were talking about for weeks. This is pretty much the largest cultural phenomenon in the US in a very long time.
Meanwhile, most normal people do not know who Sam Altman is, and a minority of them have even heard of ChatGPT.
One issue manifold has always had is with conspiracy theories - people resolving their own markets gets you a bunch of markets like "did the CIA kill JFK" where you're basically betting on how credible the market creator thinks they are, and there's always been an unfortunate amount of those on the site. But with the Gaza stuff in the news it's exploded, with a bunch of highly-active users pushing every sort of antisemitic conspiracy theory under the sun (stuff like "well obviously AIPAC/the IDF/the CIA planted/made up all the evidence for X so this market should resolve no").
Makes it both hard to trust (you now have to look up even market creators of factual-seeming markets to make sure they're not going to misresolve due to conspiracy theories) and makes it a pretty unpleasant experience to go on (especially if you read the comments, you have to wade through a lot of "the *real* terrorists aren't the rape/torture/murder advocates they're the people fighting them).
The graph says that Holden Karnofsky is at 50% extinction risk, which I think is based on an incorrect interpretation of the source. The linked source is this post https://www.cold-takes.com/some-additional-detail-on-what-i-mean-by-most-important-century/ which gives 50/50 odds for "civilization could either end entirely, or change so dramatically that 'humans as we know them today' would either not exist anymore, or would at least be a very small part of the population [and] This century is our chance to shape just how this happens".
That's entirely consistent with presently alive biological humans surviving. Biological humans could become a small part of the total population just via the creation of tons of digital minds. (Which humanity may freely choose to do.) So 50% is an upper-bound on extinction risk.
Edit: The 50% is also conditional on PASTA being developed.
Re Argentinian economy, there is a difference between Dollarized and Fully Dollarize where they try to Dollarize but it doesn't quite take and they end up with a mix of dollars and the old currency (pesos)
Manifold wants the dollar de jure in place, Metaculus wants the peso de facto gone
Thanks; I've added a link to this comment into the text.
Happy to. You could also interpret the spread as "Milei tries to Dollarize Argentina but it doesn't work and/or he is replaced by somebody who rolls it back", which I think is quite possible
What do you call it when the currency is pegged to the dollar at a fixed exchange rate?
A currency peg. They work as long as the economy is stable and the government is trusted, because they rely on the central bank buying and selling currencies to keep the actual rate equal to the posted rate, if they fail to do so there will eventually be a black market at the actual exchange rate and the peg will fail.
Author of the Metaculus question here. I believe this is correct.
Taylor Swift has been a big pop star for a while, but now she's a billionaire pop star.
https://www.foxbusiness.com/entertainment/taylor-swift-passes-billionaire-status-music-industry-era-artists-losing-millions
Also, smacking Hollywood by bypassing them to put her concert film in theaters and grabbing some of the NFL's spotlight makes it a better year for her.
Also, she has reached the level of being called out by name in a regional Federal Reserve Bank report…
«Despite the slowing recovery in tourism in the region overall, one contact highlighted that May was the strongest month for hotel revenue in Philadelphia since the onset of the pandemic, in large part due to an influx of guests for the Taylor Swift concerts in the city»
Something has definitely happened to Taylor Swift over the past year that has taken her from being part of the cultural background noise, easily ignored, to a huge player that you have to work hard to ignore. Check Google Trends if you don't believe me, there was a spike last October, and then a steady climb from March 2023 to current levels where she's a more popular search term worldwide than "sushi".
I can only attribute this to the fact that (a) she's touring and (b) she apparently has the best PR people in the world. At some point earlier this year I noticed that articles on totally unrelated subjects started talking about Taylor Swift for some reason, and it's only got worse since then.
Obligatory link to the best analysis of the TS phenomenon, for some weird values of "best": https://samkriss.substack.com/p/taylor-swift-does-not-exist
I love Sam Kriss so much.
What on Earth did I just read
I don't know how much of it to believe... it almost seems like a fanfiction of reality, but the Kerfuffle stuff is too notorious for this story to have been invented whole cloth, i think...
but gosh, what a read. thank you so much for the link
That's definitely Kriss' style. Somewhere between 50%-75% of what he writes is poignant schizophrenia.
If you enjoyed that one, don't miss his previous exploit, called "Live from the hate march", on the same substack.
I appreciate the expression "fanfiction of reality". I can almost see the snarky headline on some normie blog, "Internet rationalists discover the existence of literature - now called fanfiction of reality".
Is Kerfuffle real? I'm used to Sam making things up and Googled it to see if I could find anything about it, but I didn't.
When I was reading it, it tripped some memories that I had of something that seemed like it might fit the bill, something about an indie zine owner getting in huge trouble, but in retrospect I might have been Mandela Effecting myself
tl;dr: Swift never sexualized herself unlike Britney Spears, Katy Perry, or Miley Cyrus.
He then goes off on a bizarre tangent about Taylor Swift's navel and how he thought she was Lilith.
IMHO:
I actually find the 'never showed off that much skin' theory credible: mild feminism and attractiveness without excessive sexualization would allow her to appeal to the center-left and center-right respectively, creating a bipartisan audience that rarely exists anymore. But *that* spin is mine.
Hanania had a few posts about pop music that were interesting to me as someone who stopped following pop music around the same time as him (early 00s).
https://www.richardhanania.com/p/changes-in-music-1998-2022
He argued that white female pop stars have for the most part aggressively *de-sexualized* (i.e. androgenized) themselves while at the same time not really singing what we would recognize as a "love song" anymore.
To the extent this is true (and I'm too ignorant of pop music to really know), we might say Taylor Swift is the last pop star. She sings romantic love songs (and break-up songs, but ones that still hold romance and love in high regard) while looking pretty but not pornographic, and she does it highly competently. There used to be a lot of competition in this space, but perhaps now there's basically none?
I think I read that. I don't know enough about pop music anymore to say. Sounds credible. Hanania's a smart guy, even if he says weird stuff sometimes. Maybe she identified an underserved market and went for it.
I'm skeptical on the last sentence. I think it's exceptionally rare for musicians to pivot hard in a musical direction that goes against their instincts in pursuit of commercial success, and to then succeed for any length of time. I'm aware of two credible alleged examples: the bands Bush and Sugar Ray. In both cases, they didn't really last long.
There's a documentary about Kenny G, and it's funny because they interview all these "jazz experts" and they all hold him in contempt, describing him as a complete sellout who saw the market opportunity from watering down jazz music to make it palatable to the masses. And they interview Kenny and he basically says, "I'm really not that smart. I just make the kind of music I like to make." I believe him.
No. This is completely wrong. Billie Eilish, Olivia Rodrigo, Ariana Grande, I can keep listing names but I can say if your summation is correct Hanania is ludicrously wrong again. Like, the main most popular new white female pop stars still sing mostly about romance and look conventionally attractive.
Billie Eilish - Richard describes her as "going out of her way to look less pretty than she otherwise could" and based on an image search this seems apt to me. I also didn't think she really sung about romance; based on her debut album cover art and some vague things I heard, I thought she was positioning herself more like a young female Marilyn Manson, but I suppose I'm wrong there. But I look at her album tracklist (and again, that cover art) and see titles like "all the good girls go to hell" and "wish you were gay" and we're clearly in very different territory from Taylor or any conventional romantic songwriter of the past.
Ariana Grande - Maybe she bounces around stylistically, but I Googled her top songs and the first one that popped up was called "7 Rings". If you look at the lyrics and music video, it clearly falls into the "bad bitch" style that Richard describes.
Olivia Rodrigo - I hadn't heard of her, maybe a more promising example than the others. Though she's apparently half-Filipino, doesn't qualify as "white" in the US.
EDIT: Just looked at Richard's piece again and he DID mention Rodrigo as an example of someone who produces a new-style breakup song that is indicative of helplessness and possible mental illness, as opposed to the breakup songs of the past. Or, we might say, those of Taylor.
*Somebody* has been studying his Gonzo Journalism. The spirit of the Hunter rises again.
Yep. I don't actively look for anything TSwift related, but it seems like, in terms of career success, she's gone from "probably the most successful female artist of this generation" to "almost certainly a top 2 most successful artist ever" (alongside The Beatles).
(Also, Scott says, "she's been successful for the past 10 years as well" but frankly she's only become more successful over time *and* entirely changed genres, despite already being quite popular in 2008, which is actually quite rare among musicians. Who else was big in 2010? What are they doing now?)
Isn't there a strong case for Beyonce being nearby?
I haven't checked carefully but I suspect if you look at the numbers and the records that Swift set, it's not very close.
Sit down, Kayne.
/s
😁
Imma let you finish...
As for albums sold, Chartmaster puts Swift in 16th place. Above Mariah Carey, but below Rod Stewart. However she's moving up fast, and it's expected she'll knock Celine Dione out of 14th place soon. The only solo female artist above Dione and Swift is Madonna. She's in 5th place.
https://chartmasters.org/best-selling-artists-of-all-time/#:~:text=The%20Beatles%20are%20the%20best,solo%20artist%20of%20all%2Dtime%3F
I'm not convinced lists like this do a good job of accounting for changes in how music is consumed over time. I know they make an attempt, but even the "selling" in the phrase "best selling artist" isn't really accurate when so much is available for free. I'm also not convinced that the way they account for international developments is very good, since they rely on a separate organization in each country, each of which might do things differently.
Different lists aren't even consistent. For example https://en.wikipedia.org/wiki/List_of_best-selling_music_artists has Swift in 9th (counting by "total certified units) with Madonna (who's been active for far longer) and Rihanna above for female artists, but this metric would have Rihanna as by far the most successful artist ever. Mainly I wanted to link it because it also gives numbers for "claimed sales" which vary wildly within each artist (when multiple numbers are given), differ dramatically from "total certified units", and are only presented with very approximate values, highlighting the difficulty of measurement in this category.
I agree with you for the most part. Also, remember that this definitely doesn't reflect the size of the fanbase of any given artist at the time of this writing (2023). Many Elvis fans are now passing away. And Beatles fans are in aging and dying too. I doubt that either the Elvis or the Beatles are enlarging their fan base faster than their fans are dying off. Heck, it's the 40th Anniversary of Michael Jackson's Thriller! His fanbase must be on a general downward trend now.
Using Google's ngram viewer, it looks like the Beatles will drop from cultural consciousness sometime between 2070 and 2080, and Elvis, though he peaked higher than the Beatles is dropping faster and will probably be extinguished from our cultural consciousness in by 2060. Unfortunately, ngram viewer only tracks word frequencies up to 2019, so we cannot really see where she stands today against other pop music icons.
https://books.google.com/ngrams/graph?content=Beatles%2CElvis%2C%22Taylor+Swift%22%2C%22Lady+Gaga%22&year_start=1950&year_end=2023&corpus=en-2019&smoothing=3
As the global population continues to expand, you should look at fans per capita. Wouldn't know where to find those statistics.
Heh. Every generation rediscovers the power of branding, and is surprised at how powerful an effect it has on the popular culture. She may be this generation's Oprah.
Yep, she's exploded this year. There is a reporter dedicated to covering her, Outlets are discussing how much GDP she produces while in your, Freddie and others have started ranting on how she and others cannot be criticized. My Father in his 60s has started calling himself a Swifie and gushes over how cute a couple she makes with the NFL star. She's the biggest pop star since Michael Jackson, who was also a Time Person of the Year.
She's basically at Madonna or MJ level now. Before this year I sorta lumped her into a group with e.g. Lady Gaga and Katy Perry, as female pop singers who I knew were famous but I rarely see in the wild due to my apathy for Top40 music, broadcast TV and celebrity gossip. But she's clearly peaked well above the level of other similar artists, for whatever reason.
Re-recording several albums is nutso, shouldn't have worked, and definitely couldn't have worked in prior eras of music, but apparently did work due to harnessing absurd levels of fan loyalty and manipulating the streaming services. The well-timed fake romance with Travis Kelce, where everyone very obviously avoided saying anything untrue while still giving the impression they were dating, had to involve tons of media people playing along, but they did since it was in their own self-interest to do so. I don't know how much she herself was a driver of these recent moves, but it was rather clever. I think MJ's music was more ubiquitous, I still can't name more than 3 of Swift's songs, but in a fractured media landscape where she can't put her videos on Fox before the Simpsons and draw half the TVs in America, she's gotta be close to the peak stardom possible.
> Re-recording several albums is nutso, shouldn't have worked, and definitely couldn't have worked in prior eras of music
What about it is nutso, or unusual?
Seems to me that most performers routinely re-record their studio releases by doing live performances. A live album is mostly just re-recordings of earlier work.
Re-recording studio versions of an album and then selling it again is unheard of. Live versions are quite different for rock acts, maybe less so for pop acts but still offering something distinct.
This as I understand it was done only to own the songs, and clearly depends on Spotify and iTunes etc to offer the new version without any artistic justification for it. Whatever company now owns the original “1989” and other records can still license and sell the music, so the entire thing relies on Swift convincing people only to stream and purchase the version she owns, out of fan loyalty or antipathy to big business (as if she herself wasn’t basically just a giant corporate blob at this point.)
My understanding is that TS still holds the songwriter copyright, and is using that to prevent the original catalogue from being played. The story I heard is that her original recording contract held the copyright for the recordings. The label sold them to someone (not sure of what the history there is, but my understanding is playing TS's "Bad Blood" will provide proper mood) for $900,000,000. Of which TS received $0. So withholding songwriter permission for performances lets TS zero out the value of that property, while recording the "Taylor Version" gives her a steady income stream.
Wasn't the label itself sold to someone Swift has accused of bullying? And possibly for mere $300 000 000 – $400 000 000.
Apparently Swift cannot block exploitation of the old recording via copyright means. She seems to have enough consumer loyalty to cause a consumer boycott of those old versions though.
> So withholding songwriter permission for performances lets TS zero out the value of that property, while recording the "Taylor Version" gives her a steady income stream.
(EDIT - this following paragraph is wrong; Taylor Swift herself always held the publishing rights to her songs, and the royalties owed on the covers are therefore owed to her. That said, the value of the originals is clearly very high, not zero.)
But the Taylor's Version raises the value of the original back above zero, because she has to pay royalties when her covers play.
Also she doesn't appear to have blocked the originals: here's Taylor's Version of You Belong With Me ( https://www.amazon.com/dp/B08W92DXXM/ ); here's the original ( https://www.amazon.com/dp/B006WWU9Y2/ ); here's the original but Platinum Edition ( https://www.amazon.com/dp/B00AFDDB4Y/ ).
Others have re-recorded their albums (and/or just individual songs), sometimes for basically the same reasons, other times because they wanted to do it differently or better.
https://americansongwriter.com/4-earlier-artists-who-pulled-a-taylor-swift-and-re-recorded-their-music/
https://rateyourmusic.com/list/conchobhar/play-it-again-artists-who-re-recorded-their-own-albums/
In most of those cases it does appear there was an artistic reason cited or an obvious stylistic change. I still think it's unique to the Swift situation that she was both able to re-record an entire back catalog for purely financial reasons AND that the current music distribution landscape is such that this maneuver actually works. The Everly Brothers and Sinatra examples are cases where they re-recorded for new compilations, which makes sense in the days of physical media that you had to be offering a new product to sell. But when everybody just streams music, they're accessing it anew each time they want to listen rather than keeping a pre-existing already purchased music library, so she could lean on Spotify etc to replace the old versions with the ones she owned.
As an analogy, imagine Garth Brooks had tried this 20 years ago with his first 7 albums. There were already too many copies of those CDs in his fans' hands to get them to re-buy all of them, best he could've done was slap them all into a box set and hope to sell enough to fans that didn't already own them, but that wouldn't have justified the expense. Only in the current age, and only if you can implicitly threaten to unleash your army of drone fans like Swift can, does any of this work.
A live performance is very much not the same thing as a re-recording; recording generally involves multiple takes and layered parts. Performing an album live takes as long as the album runs. Re-recording an album can take months.
If one considers amount of news coverage, I definitely consider her in the running. By that metric, considering the competition by year she's definitely had a bigger year than previously.
I have a couple of fans in my family.
Swift has retrieved the rights to a bunch of her old songs, re-recorded them and released new versions as albums. She is also touring extensively this year.
She was well-known and famous before, but these factors have combined into some sort of next-level fame.
> Swift has retrieved the rights to a bunch of her old songs, re-recorded them and released new versions as albums.
I thought the point of re-recording them was that she didn't have the rights to the old songs? If she owned the old recordings, what would be the purpose of re-recording them?
https://www.vox.com/culture/22278732/taylor-swift-re-recording-1989-speak-now-enchanted-mine-master-rights-scooter-braun seems to make it pretty explicit that the point of the new recordings is that she owns the rights to them and she doesn't own the rights to the old ones.
Oops you're correct.
I guess the reason for the new fame that the fans are taking her side on the dispute, which is why they're buying the new versions in more than the usual numbers that would happen for e.g. a remix album.
She failed to retrieve rights to the specific recordings, but recovered rights to make new recordings of the same songs.
Perhaps, unlike so many other pop stars, she's actually intelligent and a skilled business strategist?
I mean, she's obviously cute, a good singer, and charismatic. But if she's smart on top of that...
I remember reading long ago that TS as a kid studied the downfalls of musician's in behind the music with the express goal of avoiding their failures.
Really.
That would make so much sense.
I also note her dad is an ex-stockbroker and her mom was a mutual fund executive before being a homemaker, so evidently she has genes for business. Her parents may have taught her something too.
> but recovered rights to make new recordings of the same songs.
Huh? Everyone has the right to cover a song. It isn't possible not to have that right. You can't "recover" what you never lost.
You need a license to record the cover, but issuance of the license is mandatory (and doesn't involve talking to the copyright holder) and the fee is set by law.
I think she has made sure that the cover fee for those albums goes to herself (as the songwriter)
That appears to be correct. As far as I can tell (from the wikipedia article someone linked, https://en.wikipedia.org/wiki/Taylor_Swift_masters_dispute ), she always held the publishing rights.
In reliability-of-wikipedia news, however, this claim from the same article is definitely false:
> Swift would not have been able to re-record her musical work had she not been a songwriter.[16][3]
Two citations on an obviously untrue statement. Wikipedia is self-correcting!
Taylor Swift is dating Travis Kelce. Romance involving two attractive superstars is PR gold.
This is obvious to virtually every teenage girl in the US and much of the rest of the world.
Yeah, I might look dumb in a few weeks for saying this, but the fact that the prediction markets are giving Altman a higher chance of being POTY than Swift doesn't speak well of them, imo.
Semi-snarky nit: Should ChatGPT really be in the running for _person_ of the year? This feels like jumping the gun on claiming to have AGI to me... Or does passing the Turing test (with whom as the judge?) suffice?
Maybe not, but Time one year did "You, yes you, the reader" as their "person of the year" so it wasn't inconceivable.
Good point!
Well, that was the easiest $150 I ever made
Well done, though I don't think this was as easy a call as you say. They've never had a pop star or for that matter any kind of entertainer as person of the year. They've almost all been some sort of political, social, religious or business figures.
Yeah, I would've been less confident if I'd realized an entertainer had never won it, but 1. I think she was clearly the most likely winner on the shortlist (Hollywood strikers, Xi Jinping, TSwift, Altman, Trump Prosecutors, Barbie, Putin, Jerome Powell), and 2. Having Altman at twice her price was just insane, so it was pretty clear to me she was undervalued. Time is in the business of selling magazines, and if they'd put Sam Altman on the cover, 90% of people would've said, "Who?"
Yes 2023 was massive for Taylor Swift:
- highest grossing tour of all-time https://www.billboard.com/pro/taylor-swift-eras-tour-top-grossing-global-tour/
- highest grossing concert film of all-time https://deadline.com/gallery/highest-grossing-concert-movies-all-time/taylor-swift-the-eras-tour-seattle-wa-2/
- released re-recording of one of her highest-selling albums
- was #1 streaming artist on Spotify (was #2 the last two years and not in the top the years before that) https://en.wikipedia.org/wiki/List_of_most-streamed_artists_on_Spotify
- major topic of sports news / celebrity gossip with Travis Kelce
- became a billionaire
That said, TIME has never chosen a musical artist for POTY.
Interesting how an entire thread on the world's biggest pop star at the moment doesn't seem to include a single word about whether her music is any good. I take it as an indication it might be not - but maybe I'll have to do the supreme effort of firing up youtube and checking for myself
OK, had a good listen around her albums. Not exactly my cup of chai, but she's clearly a good artist. Excellent production, good songwriting, expressive voice, good grasp of the zeitgeist, variety of styles, and weaving a fine line between emotionally engaging, assertive but also kinda harmless. Lots of moody minor chords, tonal ambiguity and electronic textures. "Red" is probably the best of what I've sampled - but I still prefer the King Crimson album of the same name.
Ha! I fear that opinions about the quality of her music will be more subjective than discussing stats about her reach. Personally I’m fairly indifferent about her, but she certainly seems to deeply resonate with a large number of people. I suspect it has something to do with the fact that she writes her own songs - not that common in pop world - and she combines her talent with a vulnerability and openness that seems to connect with fans who like to dissect the meanings of her lyrics (even if it’s as apparently shallow as “which ex is this one about?”) and they end up feeling like they know her deeply or something like that. Rumor has it she’s also got some business savvy.
The Eras Tour was the highest grossing tour in U.S. history. It generated small booms in excess consumer spending in every city where she performed; According to current estimates, about $4.6 billion in excess consumer spending. It's like Swift ran her own private stimulus program for post-COVID downtowns. If you worked in or near a city she performed in it was all people were talking about for weeks. This is pretty much the largest cultural phenomenon in the US in years and years.
How is a commitment to merge with other labs close to superintelligence even plausibly a good thing to commit to if you are concerned about alignment? What if those other labs have different ideas about alignment? What if one of them actually wants an unaligned AI?
I mean either it's essentially meaningless or it's dangerous. If it only requires you merge if the other labs acceed to putting you in charge or at least adopting the approach you prefer then it's just kinda silly. I think pretty much any CEO and board of a non-profit will always accept some other organization with a bunch of assets showing up and saying: you can direct all our resources too.
OTOH if it means serious negotions to work out a power sharing agreement that induces other institutions to accept then it obligates you to share power with groups whose goals (or at least methods) are directly opposed to your own. And if the then board gets to decide if any other institution is too incompatible it's really saying: merge if it seems wise to do so which is the same as not saying anything.
--
My concern isn't the legal enforceability. Often it's good to put in commitments to a charter that lack any realistic enforcement mechanism. Rather, it's that either good faith board members are either agreeing to a rule they should know would be quite harmful if they felt obliged by it or it's misleading to phrase it as a commitment and should really be more like: when you get close stop and do whatever helps.
You're close to ASI, and have competitors who are also closing in. There is the fog of war. Would you prefer to rush alignment so you can be first (increasing risk), or try and negotiate with the other labs so all of you can actually work out alignment?
The charter says "if a value-aligned, safety-conscious project comes close to building AGI before we do, we commit to stop competing with and start assisting this project." I think the theory at the time was that a major issue wouldn't just be solving alignment, it would be making sure that the first group to ASI would have enough of a lead to be careful about implementing it, and wouldn't rush the last steps because of uncertainty about less-careful competitors.
Ok, thank you. That explains the thought process. I was just thinking about it quite differently.
It does basically commit openAI to scuttle any attempt at a pause and seems to ignore the possibility that the first group to get close might be motivated by other concerns than capitalistic ones -- eg it won't be people who are pursuing the project because they believe it will save their country from some kind of war or political crisis.
> a graph of different people’s probability estimates of AI causing human extinction
Are these probabilities of extinction conditional on AGI or extinction in general? There's plenty of future outcomes where AGI is never built for one reason or another.
What are some of those? Absent AGI, or similarly advanced tech, there are no plausible total extinction scenarios in the foreseeable future. Technological progress is a robust attractor that civilization would eventually converge to even if the current iteration suffers some catastrophic disruption like a global thermonuclear war, which would once again put it on the path to AGI, another robust attractor, and seemingly in easier reach than Drexlerian nanotechnology or brain emulation.
You don't need full nanotech to end the world; an alga that can't be digested, has a better alternative to RuBisCO and needs 1000x less phosphate is enough (because it blooms like crazy across the ocean and ties up all the atmospheric and then biospheric carbon as gunk on the seafloor). This is *mostly* a computational problem at this point, although the assembly would also be mildly beyond current SotA.
Redirecting a sufficiently-large asteroid into Earth would do the trick if self-sustaining offworld colonies did not yet exist. Chicxulub wouldn't be enough, Ceres would, not sure about the actual crossover point.
Scaling up particle colliders enough could lead to micro black holes.
The amount of dakka required for these is large, but they don't really need super-tech.
All of those seem to be deliberately self-destructive while requiring non-trivial effort, which, to me, puts them below the level of consideration. After all, we could always have just murdered everyone with sticks and stones, but somehow have managed to avoid doing that for millions of years. Meanwhile, stuff like AGI and full nanotech are dangerous precisely because of their vast non-destructive potential, but which could catastrophically spiral out of control.
I agree that in general it's not worth giving these scenarios much consideration, but I don't think I agree with your reasoning. Individual human beings do, indeed, murder people with sticks and stones. They're a small proportion of the population, though, and they can't kill more than a handful solo before someone stops them. the probability of so many individual humans all committing a bunch of murders all at once is vanishingly small.
I have no doubt that there are individual human beings who really existed who, if given the opportunity, would have killed more but lacked the capability to do so.
once nukes were on the scene, the bar for human-caused extinction finally came within line of sight of our species, but even getting close still required the resources of a major national government. as the bar lowers, the number of simultaneous crazy-psycho events needed to seriously screw things up drops
as technology advances, that bar gets lower, eventually it gets so low that, idk, a random dealthcult with a single microbiology PhD could pull it off. if we survive that, eventually it gets so low that a single person with a garage bioreactor could do it.
I would not be so quick to write off 'obvious malicious self-destruction' timelines
>Individual human beings do, indeed, murder people with sticks and stones.
Overwhelmingly pursuing their self-interest. There were a few moderately "successful" suicide cults over the centuries, but I wouldn't expect their future analogues to get world-destructive capabilities before actually powerful actors are in position to destroy the world seeking beneficial ends.
The first one will be sold as «highly efficient biological carbon capture». Won't even be a lie. No need to reveal (or notice in time) that your breakthrough technology just happens to have side effects.
That's not even a side-effect, it's just overshooting the goal.
The particle collider and the Life 2.0 can happen without actual malice; we have GoF research right now despite it being -EV and despite the people doing it not generally wanting to kill millions with a pandemic.
The asteroid does need malice, but it's plausible that someone living off-world might not understand that his/her colony is not self-sufficient and thus it only absolutely requires war malice, not omnicide malice.
I do agree that the continued GoF clown show is some evidence that humanity could sleepwalk into self-destruction without actual malice, but I'd still in general expect several progressively worse non-total-extinction warning shots due to such brainfarts.
Pandemics and asteroids yes, Life 2.0 and particle collider no. A particle collider either produces ice-nine of some sort (black holes, stable strangelets, true vacuum, etc.) or it doesn't; there's no in-between. Life 2.0's not quite as obvious - bad-but-not-X Life 2.0 is physically possible - but the things that would kill us are almost the easiest things to get working at all.
Anything that would cause a serious WWIII. One crazy politician in the right place might be sufficient.
WWIII isn't an X-risk. GCR, definitely, but not X. Direct blast/heat/etc. can't kill rural communities (or towns/cities not nuked, and there aren't that many nukes anymore), fallout decays to tolerable levels in weeks, and nuclear winter a) is mostly a hoax, b) would have to be hilariously, unrealistically bad to actually finish us all (as even hunting remaining possible would preserve the species, so the entire ocean would need to freeze over).
Both biotech and nanotech are possible X-risks. Diseases are not designed to kill people but if the technology for designing them gets good enough they could be. Grey Goo scenarios with Drexlerian nanotech are possible. Either could conceivably wipe out our species.
It seems like these AI experts have a blind spot in that they are unable to consider the likelihood that AI could be beneficial to humanity. When discussing a range of probabilities, if there's a chance that AI could cause human extinction, what's the chance that the AI could be extinction-neutral? — or could *prevent* human extinction? I can just as easily imagine a Banksian Culture-like future where AI 'Minds' have endearing and eccentric personalities and get along well with humans.
I think there is a deep psychological need in many humans to indulge in apocalyptic fantasies. Since the 1970s I've been told by "experts" that humans are facing extinction by a continuously-changing set of threats—
1. Pollution (will kill the O2-producing algae)
2. Club of Rome prediction that civilization will collapse by the year 2000
3. Global cooling and the coming ice age
4. Nuclear winter
5. AIDs will infect one-third of humanity (ending civilization);
6. Global warming
7. Ocean acidification (will kill the O2-producing algae)
8. Asteroids
9. Y2K
10. COVID-19 will be a mass disabling event (ending civilization)
11. and now it's AI.
Those first five predictions haunted my childhood and young adulthood until I started to realize that most experts were either purposely bullshitting us and/or themselves.
Meanwhile, my Xtian neighbors kept warning me that the End Times were coming any day now!
Neither the experts nor the religious have been correct in their predictions. I just take them all with a grain of salt now. Humans are incapable of predicting their future, and it will most likely be something we didn't consider that will take us down. ;-)
"3. Global cooling and the coming ice age"
Ah, you remember that too? A good while back I got plenty of "nope, never happened, nobody ever said that, you're mistaken/lying" from the disgruntled CLIMATE CHANGE IS GONNA KILL US ALL types on here when I mentioned that the coming New Ice Age was one of the flaps of my youth.
At least if I hallucinated it, I'm not alone 😀
Oh, and don't forget: hole in the ozone layer! We were all going to get skin cancer and die from that one.
Eh, to be fair, most of these are attributable to the journalists (who have only gotten worse lately), not to the actual scientists (nor even policymakers).
1. "Pollution (will kill the O2-producing algae)": Still a real problem, only it depends on the kind of pollution. For example, fertilizer pollution creates delightful and vibrant algae blooms (that are also toxic to all other forms of life).
2. "Club of Rome prediction that civilization will collapse by the year 2000": No idea what this is, never heard of it. Sounds like a newspaper thing.
3. "Global cooling and the coming ice age": Was never a real thing, media companies invented this one.
4. "Nuclear winter": And that's the least of your problems if global thermonuclear war actually happens; and with events like the Cuban missile crisis, it seemed like a real possibility... and it still is to some extent. But for now we seem to have pulled back from the brink.
5. "AIDS will infect one-third of humanity (ending civilization)": It's kind of already the case in Africa, sadly; but the world-wide prediction was a media invention.
6. "Global warming": Would be lovely to say it's all fake, but right now looks like it's all too real. Climate skeptics have mostly switched from saying "it's a hoax" to "it's totally happening but maybe that's for the best". Hmm.
7. "Ocean acidification (will kill the O2-producing algae)": Not the algae (they have a pretty quick evolutionary cycle), but certainly the coral reefs. If you wanted to see one in all its glory, now's probably your last chance.
8. "Asteroids": They're a real though relatively unlikely threat. Have you seen the Chicxulub crater ?
9. "Y2K": This one was a combination of real problem that was successfully solved by a concerted effort, and OMG media overhype to the max. If we did nothing, chances are that e.g. banking systems would crash, which would be... not great.
10. "COVID-19": Could've been much worse without MRNA vaccines, but like AIDS, somewhat overhyped.
11. "AI": What pisses me off here is that all the Singularity hype is masking the real danger. There are people now who are cheerfully letting ChatGPT write their laws or even their software. This is like putting a typewriter monkey in charge of same.
12: "Ozone Layer": This was a very credible threat which was solved through a concerted effort (banning CFCs). For once, humans did something right... sadly we probably lost the capacity for such action by now.
"Was never a real thing, media companies invented this one."
You mean they used speculation to write up scare stories? Say it ain't so!
https://en.wikipedia.org/wiki/Global_cooling
"Concern peaked in the early 1970s, though "the possibility of anthropogenic warming dominated the peer-reviewed literature even then" (a cooling period began in 1945, and two decades of a cooling trend suggested a trough had been reached after several decades of warming). This peaking concern is partially attributable to the fact much less was then known about world climate and causes of ice ages. Climate scientists were aware that predictions based on this trend were not possible - because the trend was poorly studied and not understood (for example see reference[16]). Despite that, in the popular press the possibility of cooling was reported generally without the caveats present in the scientific reports, and "unusually severe winters in Asia and parts of North America in 1972 and 1973 ... pushed the issue into the public consciousness".
Well, tell that to my (late) Glacial and Quarternary Geology professor. He had graphs and data showing that we were entering the cooling phase of the current Milankovitch Cycle. It was taken very seriously despite what Wikipedia might say. Suggest at John Imbrie's work from the 1970s and 80s.
"2. "Club of Rome prediction that civilization will collapse by the year 2000": No idea what this is, never heard of it. Sounds like a newspaper thing."
There was a book version too: https://www.clubofrome.org/publication/the-limits-to-growth/
Scathing review in https://www.americanscientist.org/article/computation-and-the-human-predicament
"In this context climate models offer a useful point of reference. General circulation models for the atmosphere and the oceans, along with related models of ice sheets and atmospheric chemistry, have several points of similarity with World3. At a conceptual level the structure is much the same: There are flows of air, water, heat and other entities, which the model must sum or integrate. The time scales are similar: In both cases we want to know what’s going to happen several decades out. And feedback loops are essential mechanisms in both kinds of models. (There are even historical connections. The use of general circulation models to study global climate change began in earnest at MIT circa 1970. The instigator was Carroll Wilson of the Sloan School of Management, who was also the person who got Forrester involved with the Club of Rome.)
These similarities are outweighed by differences. Where the Limits team had a casual attitude to data gathering practices—and outright hostility to statistical methods—the climate science community is passionate about collecting data, verifying its provenance and quantifying its uncertainty. General circulation models are not based on rough estimates or guesses but on decades of meticulously curated measurements—what Paul Edwards, in A Vast Machine , calls a “climate knowledge infrastructure.” "
The other big difference is that _Limits to Growth_ ignored rational human responses. In their model, if food prices went up, output went up in the short run then down as farmers destroyed their soil trying to produce more. That assumed that the farmers ignored the fact that high prices gave them an incentive to maintain the soil for future output. As I put it at the time, the model was like predicting the path of an automobile whose driver couldn't see through the windshield.
Climate science doesn't have that problem as long as it is only dealing with climate. It only comes up when it is being used to predict the effects of climate change, which isn't really climate science. The amount of CO2 depends on human action but the ICCP can and does deal with that problem by considering multiple emission scenarios.
Many Thanks! Yes, I'd heard that criticism too.
One more quote from the review: "As for the mathematical model behind the book, I believe it is more a polemical tool than a scientific instrument."
1. Ocean anoxic events (OAEs) have happened in the past (probably significantly to the Permian mass extinctions), but the claim that pollution and/or acidification will cause an OAE event seems pretty far out on the range of probabilities. Especially since we've significantly cut down on our water-polluting ways since the 1970s—and because the evolution of plankton during the Jurassic that sequestered carbon in calciferous sediments will buffer acidification in the longterm.
https://www.quantamagazine.org/how-jurassic-plankton-stole-control-of-the-oceans-chemistry-20191001/
2. Club of Rome and the famous/infamous publication "The Limits to Growth". ITheir report kicked off a whole slew of books about the uncertain future of humanity...
https://en.wikipedia.org/wiki/Club_of_Rome
https://en.wikipedia.org/wiki/The_Limits_to_Growth
3. Global Cooling was a real worry. If glaciation is dictated by Milankovitch cycles, then we're entering a period of global cooling. See the papers by John Imbrie.
https://www.science.org/doi/abs/10.1126/science.207.4434.943
Also this popular documentary with our favorite Vulcan narrating.
https://youtu.be/NQSBn50o_8M
4. Carl Sagan and nuclear winter. Considering there were over 500 above-ground nuclear tests during the 1950s, do we have any data that suggests that they affected the climate? I don't know. I haven't looked into that question.
https://www.sciencefriday.com/segments/nuclear-winter-carl-sagan/#:~:text=Carl%20Sagan%2C%20writing%20for%20Parade,the%20general%20public%20about%20it.
5. "AIDS will infect one-third of humanity (ending civilization)". Can't find the links because they're pre-Internet, but I think Douglas Feldman was sounding the alarums about extremely high mortality rates from AIDs. This was before they really had a handle on its R0, though. I remember Scientific American amplified his concerns (fears).
6. I'm afraid I'm pro-global warming. Until recently CO2 levels were the lowest they've been since the Permian. Angiosperms—which are responsible for most of our food supply—evolved when CO2 levels were 4x-5x higher than they are today. We can see a positive response to higher CO2 levels to the current greening of the world.
https://www.nasa.gov/technology/carbon-dioxide-fertilization-greening-earth-study-finds/
7. Coral reefs? Coral reefs existed when the global climate was an average of 6º C warmer and CO2 levels were 4x higher than today.
https://royalsocietypublishing.org/doi/10.1098/rsta.2013.0096
Skip to 12. I hate to tell you but the Ozone hole is still opening up every year over the Antarctic. 2023's ozone hole briefly opened enough to be 12th largest hole since they started monitoring it—but overall, 2016 was the 16th worst from Sept. 7 to Oct. 13...
https://www.nasa.gov/earth/climate-change/ozone-layer/2023-ozone-hole-ranks-16th-largest/#:~:text=The%202023%20Antarctic%20ozone%20hole,day%20ozone%20hole%20since%201979.
How could I have forgotten about the Ozone Layer?! There's still a big hole that opens up over the South Pole every year, It was the 12th largest on record in 2023.
I was taking classes in Glacial and Quaternary Geology when the narrative transitioned from a consensus of global cooling to a consensus of global warming. It happened within a few months by a few high-profile think pieces in the Scientific American and Science magazine (and maybe Nature, but I couldn't afford to subscribe to Nature at the time).
This documentary has survived the cultural memory loss...
https://youtu.be/NQSBn50o_8M
> It seems like these AI experts have a blind spot in that they are unable to consider the likelihood that AI could be beneficial to humanity. When discussing a range of probabilities, if there's a chance that AI could cause human extinction, what's the chance that the AI could be extinction-neutral?
Isn't the answer, according to the chart, that AI experts think AI will most likely be beneficial or extinction neutral? There are only a few people there who put the odds of extinction from AI above 50%
Note that The Precipice, essentially the X-risk Bible by Toby Ord, put AI X-risk as bigger than all other X-risks put together for the next century.
Yes, AI could help with other X-risks, but without alignment the cure's worse than the disease.
I think there are a couple positions on how to balance AI benefits vs. risks:
1. AI would be incredibly beneficial to humanity if we got it right, but there's no chance we will, so let's focus on not dying.
2. AI would be incredibly beneficial to humanity if we got it right, but better to take time to do it right and wait an extra 10 years for the benefits than eat a 20% risk of dying.
3. AI would be incredibly beneficial to humanity if we got it right, but we should also work on controlling it, for the same reason that even though planes are beneficial, we work on ways to prevent them from crashing.
Given that people have written whole books on these things and balancing the benefits and risks, I think it's unfair to call it a "blind spot".
Also, you're doing a thing I should give a fallacy name to and write a post about, which is giving a long list of things that didn't happen, then using it to prove nothing ever happens, without also including a list of things that did happen:
- The Black Plague
- The Mongols killing 10% of the human population
- Two World Wars
- Global warming of the type we've actually gotten
- AIDS killed however many people it actually killed
- COVID killing however many people it actually killed
Before all these things (except maybe the Plague and the Mongols, I don't know enough history to know what people were saying there), people were saying it was alarmist and would never happen, and those people were wrong too. I think the probability (something bad happens|people predicted something bad would happen) is moderately high and definitely not zero.
I admit humankind has never been destroyed before, but that's a tautology - it's logically necessary that the day before humankind is destroyed, humankind will never have been destroyed before!
> I think there are a couple positions on how to balance AI benefits vs. risks...
As usual, I'd like to point out that alternative positions do exist, i.e.
4. AI could potentially be incredibly beneficial (or possibly harmful) to humanity, but its current iterations fall laughably short of that goal, and there is as of yet no clear path from here to there. We should continue working on AI research, but meanwhile let's make sure we use existing (and incredibly unintelligent) machine learning systems responsibly and without undue exuberance.
Yup, all 11 sound familiar.
2 quibbles:
a) Re 8. Asteroids was always a low-probability-per-year hazard. I don't _think_ I ever heard it painted as a will-kill-us-in-the-next-decade kind of risk. ( The Yellowstone caldera giving us a supervolcano eruption is similar)
b) For a _lot_ of these, there is an immense amount of noise and oversimplification injected between what researchers in the field concluded and what the pundits spouted
This is late, but for your reference, up until 5-10 years ago, I was largely pessimistic about the long-term future, because I thought we were stuck on a single planet and there were too many things we could do to end our civilization or existence. My hopes were placed in the creation of self-sustaining, self-replicating space colonies, or building something smarter than us to solve the problems we can't. But this was all centuries or millennia away, and didn't affect my day-to-day thinking.
On the first angle, Elon Musk happened, and despite his erratic behavior, I'm quite happy that we're finally moving in a positive direction again.
On the second angle, it's true that hanging out in Eliezer-Yudkowsky-adjacent spaces got me thinking about AI risk in a more serious way. But the main cause of my pessimism came from a close encounter with a malevolent human with superior social skills. The aftermath convinced me that practically no one thinks about that sort of risk clearly, with the notable exception of Eliezer-influenced folks. The only reason my p(extinction) isn't higher, is that it's blocked by a large p(total civilizational collapse) before an AI gains the capability to kill us all.
Part of the reason for Taylor Smith being so high now is that she's on Time's released-Monday-morning shortlist ( https://time.com/6341947/person-of-the-year-2023-shortlist/ ).
It's interesting to see that several candidates on the shortlist are in fact not people:
> Hollywood strikers
> Trump Prosecutors
> Barbie
It's also interesting that Putin is on the list; I would have expected a lot of government pressure not to do that.
Every once in a while the Person Of The Year is the villain instead of the hero.
"Trump Prosecutors"
Oh, if this were a fair world...
Hey, MY hair is that stylish and I’ve been using Manifold for over a year! Deleted my manifold.love profile because of privacy concerns, but if you want IRL proof re: my hair, text your shidduch resume to 234-5[gematria of my first name][taryag].
> I’m most unsure about whether Adam D’Angelo is a committed safetyist. He hasn’t said so explicitly and isn’t openly linked to the safetyist movement.
He's not particularly concerned about existential risk, especially in the near future.
Speaking of this guy, does anyone have a bit of insight on what he's even doing on the board of what is arguably the world's hottest company? As far as I can tell he's mostly known for being the CEO of Quora, a much hated site that basically SEO-spams Google with mostly crap user-generated questions and answers. I'm sure the operation makes plenty of money, but still... what gives?
I heard somewhere that putting him on the board was part of a deal where openAI got to train on Quora data, which is copious and already ranked by answer quality. Also imo lots of quora answers are pretty good.
I must not be in the target audience... most of the time I find much better info on wikipedia and random blogs, and even by asking chatgpt, despite its hallucinatory habit. And I can't stand login walls - we live in an information glut, if I have to jump through hoops to read your content, I just close the tab and look elsewhere.
Great post! Fixed the Ilya superalignment market for you. You know there's no Prime Directive here, you could just buy it up yourself!
Any chance we can get your prediction for POTY?
Thanks Scott! Manifold is now 2 years old, as of December 1st.
We've built something cool, that a lot of people love. And yet I feel there's still so much work left to do.
> Sam tried to out a board member.
Is this valid usage for other people? To me what it means is that he publicly identified the board member as gay.
I assume it's a typo/eggcorn for "oust", as in the other market.
Scott copies the usage in his first paragraph discussing it, but then switches to "oust" in the paragraph following that.
It's close enough to making sense that I'm not really comfortable declaring that it must have been an error by the question author and then another error by Scott, but I can't rule it out either.
I think it's an error by Scott, influenced by the question author's error. But to answer your original question: no, that's not a standard usage for me either, to the extent that I assumed it had to be an error.
On average American P(doom): https://forum.effectivealtruism.org/posts/Rg7h7G3KTvaYEtL55/us-public-perception-of-cais-statement-and-the-risk-of#Perceived_likelihood_of_human_extinction_from_AI says 26%, but this clearly doesn't reflect their "true beliefs" (insofar as those exist) because if you ask Americans to rank x-risks they put AI pretty low.
Normies don't have coherent worldviews
Does anyone?
Probably not 100% but some people are a lot closer than others
It all depends on how you ask the question. No one is immune from framing effects.
It's weird that the only time Lawrence Summers had anything to do with tech companies is having one meeting with the Winklevoss twins vs FB during his Harvard years. Everything else is politics, World Bank, Harvard, hedge funds and the Treasury/National Economic Council stuff.
He has, however, mentioned AI before:
> In 2018, Summers disputed the claims from then Treasury Secretary Steve Mnuchin that AI would not replace American jobs for 50 to 100 years.
> “The robots are coming," Summers wrote in the Washington Post. That year, he also warned of economic catastrophe if the US “loses its lead" in biotech and AI to China.
https://www.livemint.com/companies/larry-summers-on-ai-governance-openai-must-cooperate-with-govt-for-national-security-and-regulatory-issues-11701573927239.html
He's survived a high-profile cancellation attempt, I wonder if that's relevant?
It's not helpful to think of cancelation attempts as a binary, because Summers seems to have landed somewhere in the middle of the spectrum. Wikipedia cites this commentary on the end of his Harvard presidency and on his less prestigious role in the Obama administration (compared to the Clinton administration): https://www.independent.co.uk/news/world/americas/summers-sexism-costs-him-top-treasury-job-1033373.html
Sure, but he's already been cancelled once, which means he's more likely and able to stand up to cancellation threats in the future. That may or may not be part of their calculus.
> This has got to be a stunt by Aella, but that means it is very likely to be fun.
FTFY.
I wonder if you could make it easier for prediction markets to handle "why" questions by mixing them with Pol.is-style crowd analysis (incetivize users to submit and vote on binary statements regarding the market->Use a dimensionality reduction algorithm to find the primary dimensions of disagreement->Let the options of the market be defined by these dimensions).
That's cool! But also complicated. I think it would help a lot to just filter submitted answers through a poll of random other users to see if they think it's a good answer (clear, not a repeat, etc) and require 5/10 to say yes.
I wish them luck. It would be neat to see "why" questions usefully examined by prediction markets. Though I shudder to think how difficult and contentious the judging task is going to be...
> Are some of these people normies? Seems surprising, but I can’t rule it out.
I guess it also depends on where you draw the line? For an extreme example, some maths PhDs have many normie leanings and count as normies in some contexts, but are definitely not close to modal normies when the question is about enjoying the precise-sciences in-jokes…
Some of those profile photos are pornbots (as we call 'em over on Tumblr). I image searched a couple because they looked a little bit too identical, and one at least is an image for modelling bikinis, which is also used as a profile photo on an X account that is - I won't say OnlyFans because I don't know, but that kind of "follow me for hot naughty content" bait.
So congratulations, the dating prediction market is mainstream enough to have the bots and spammers interested! 😁
There could be separate bets on who is a bot, etc. It's a prediction market, we can solve every problem by more bets!
Wait, no. That’s a terrible idea. Scammers could make real money by spending fake money (mana) to short the fake-money price of their “is this person a bot” market. This would cause lonely nerds to lower their prior on that person being a bot, making them more likely to get scammed.
Rough rule of thumb: if the person's photo has them with their bazooms bustin' out of their tops (male or female, equally applies) and pouty vixen/saucy expression, it's a bot. Or at least, a very good chance of same.
I would argue that, by now, any profile on any dating site is most likely a bot. That should be your default assumption.
Many of the people on those profiles are genuine because they look like people who would go on a Rationalist dating site (I know I'm leaving myself wide open on that one for "So what exactly do you mean by that, huh???" but yeah. You know your people.)
To answer Scott’s Taylor Swift question: https://www.thefp.com/p/taylor-swift-unites-america.
> This will be a total unfair stereotype, and you should feel free to yell at me for it, I just think usually prediction market junkies don’t have hair as stylish as the woman in Picture #3
Ok, I’ll do it- I think this comment is unnecessary and pretty alienating. I understand that it’s lighthearted, but people feel hesitant enough already about not “fitting in” to the rationalist community that and jokes like these reinforce that. Also, putting myself in that person’s shoes- I wouldn’t like to submit a photo to a dating site and have it remarked on to a wide audience in any way, positive or negative.
On the other hand, if you're putting a photo up on a dating site, you are opening yourself up to comments by people at large, even if that's just the people also using the dating site. You will be judged by them on "ooh, love that hairstyle/oh gosh, what a raggedy mop" as to whether they'll engage further with you, and sometimes maybe you need advice as to "hey guys, not getting any matches, what am I doing wrong? here's my profile, take a look" and yes, it's "ditch that photo, it's not doing you any favours".
I agree, but I don’t think this blog post is either of those scenarios.
Yeah, I acknowledge this, but I was trying to make an important point (it seems like Manifold Love has spread to a normier demographic than I expected) and I couldn't think how else to do that. Pointing out the black guy in the hoodie would have been even worse! But I don't think I'm imagining the demographic shift, and I'm not sure how else to evidence it.
Some people have proposed that all of those accounts are spambots because they don't have a profile, but ironically the woman with the stylish hair does have a profile and seems real.
> I wouldn’t like to submit a photo to a dating site and have it remarked on to a wide audience in any way, positive or negative.
This is more or less exactly what pictures on dating sites are for. You’re intentionally broadcasting your photo to a wide audience in the hopes that it’s positively noted. There’s no way one can seriously claim there is a privacy or abuse implication to this.
Also, the implication that this person would be *upset* by having someone note that they have a more stylish appearance than the median person of their in-group, and therefore it would cause them anxiety centered around belonging to the group, is… inexplicable.
Nobody ever got turned away from a lunch table in the cafeteria for being too good-looking.
Having many people separately evaluate your appearance, for the purpose of dating, is different from having someone use your appearance to make a point in a blog post.
Yes, but not in any way that matters. I think you’re having a hard time making a specific criticism because there’s not anything for you to find. This is reaching.
> Nobody ever got turned away from a lunch table in the cafeteria for being too good-looking.
«Stylish» means not just too good-looking, it includes choices beyond taking care of the health side. And «if you come here wearing a suit, you don't really belong here» is not and unheard-of sentiment in tech communities. So «too stylish to belong» is not outlandish enough to be unthinkable as meant literally.
Of course what matters is how the actual phrase written — in the context of the disclaimers, which do exclude literal meaning — reads for people who could be plausibly meant by references to stylishness. And it matters that both «ACX readers» and «Manifold players» are skewed towards having an idea what weak evidence is. I can't be accused of being stylish in anything, so no idea how it works out at the end.
You're taking offense on behalf of your imaginings of someone you don't know because they might be alienated by a positive comment about their hair.
I'm a huge Swiftie, so I don't claim to be unbiased. But the Swift mania is off the charts more than ever this year, and with good reason. Here's the best case I can make for her as Person of the Year:
1. Eras Tour
Everything about the Eras Tour is absurd. Who performs for 3.5 hours straight with 10 different costume changes? And then does that for several nights in a row? And then repeats the spectacle for 146 nights over nearly two years? (The tour started in March 2023, and the international leg continues until December 2024!)
Her opening night brought in more business for the host city than the Super Bowl: https://time.com/6307420/taylor-swift-eras-tour-money-economy/. And that was just one of 52 shows in the US alone. Overall, her tour brought in nearly $1 billion in revenue this year, making Taylor a billionaire. And it produced around $5 billion in revenue for the broader tourism industry.
In addition, the millions of people who attempted to buy Eras Tour tickets crashed the Ticketmaster site so badly that it led to a congressional investigation. This included multiple senators dropping references to Taylor's lyrics in an attempt to seem cool and hip: https://www.cbsnews.com/news/taylor-swift-ticketmaster-senate-judiciary-committee-eras-tour/. (IMO, all of them failed as miserably as one would expect, but the obvious attempts at pandering show how influential Taylor is.)
On the side, Taylor also released an Eras Tour movie, which was one of the top-earning movies of the year. Its box office total has now passed $250 million: https://www.msn.com/en-us/music/news/box-office-milestone-taylor-swifts-eras-tour-crosses-250m-globally/ar-AA1kDfb0.
2. Disrupting the music industry
Taylor's success is due to her business savvy just as much as her songwriting ability. She's constructed a brand around herself in ways that few other artists can touch. And the cornerstone of that brand has been her brilliant moves over the past few years to gain control of her own work.
To recap, in 2019, the original masters of Taylor's first six albums were sold to a private equity firm over her strenuous objections. But Taylor found a loophole - the copyright only applied to those specific recordings. As was standard for the industry, Taylor's contract allowed her to create new recordings of her entire catalog five years after the original albums were released.
Before Taylor, no one had attempted to exploit that loophole at scale. Perhaps that's because no one had achieved much success with re-releasing songs that everyone already knew. But through a combination of adding new songs ("From the Vault") and appealing to her loyal fanbase, Taylor made it work better than most people imagined possible.
She has since re-released four of her first six albums as Taylor's Version (two of them coming in 2023), with two more likely arriving soon. All four have debuted at #1. The most recent and most successful was 1989 (Taylor's Version), which sold 1.653 million units in its first week. That's larger than any other album since 2015, and significantly larger than the original release, which opened at 1.287 million units.
In response, the major music labels have all rewritten their standard contracts to prevent future artists from replicating her success: https://www.billboard.com/pro/taylor-swift-re-recordings-labels-change-contracts/.
So it remains to be seen whether Taylor has created new opportunities for artists to take more ownership of their work, or whether she's only exploited a temporary loophole for her own advantage. But either way, she's definitely disrupted the industry.
3. Warm fuzzies
This is admittedly subjective, but the piece that Petey shared points in the same direction: https://www.thefp.com/p/taylor-swift-unites-america.
Of all the options for Person of the Year, Taylor Swift is by far the most likable. That's why she's the overwhelming favorite in the reader's choice market: https://manifold.markets/Joshua/will-taylor-swift-be-the-time-magaz.
At a time when the news is full of doom and gloom, and the country is more polarized than ever, Taylor is the one major person who is nearly impossible to hate. And that makes her the Person of the Year that we need right now.
I don't know how the Person of the Year is chosen, but surely it's better for Time to have Swift on the cover than Altman?
Better in what way? If Time says the most important person this year is a pop star who has done nothing but be really good at pop starring, that shifts their brand towards Teen Vogue celebrity gossip. If they name the head of an $86 billion tech startup promising to change the world, that's a very different brand.
I have no idea which is the better business strategy for Time, maybe the celebrity gossip market is hot right now, but I'm pretty sure the magazine's current brand is a lot closer to Altman than Swift.
My guess is they meant better in terms of the literal cover photo, as in you can get a very good-looking picture of Taylor Swift.
Quite so.
Relatively fewer people know about Sam Altman.
The purpose of the (overhyped) POTY stuff is (free) publicity for Time magazine.
Altman on the cover would have produced “who?” from too many people.
But Time's Person of the Year hasn't been serious in a long time. In 2001, they made it Giuliani rather than Osama Bin Laden. In 2003 it was "The American Soldier," in 2005 it was "You."
Which is great whenever I get into an argument with 18 year olds, I can always win with "have YOU even been Person of the Year? I didn't think so."
In 1988 they gave it to the planet Earth considered as a whole. Raising the question of why, if the entire world can collectively count as a single person for the purposes of the award, it doesn't win *every* year.
Thanks for the explanation, the swifties have remained easy to ignore still but the phenomenon is interesting.
Since you are clearly a big fan, and I barely know her music from snippets on the radio, you seem like the guy to ask this, because I'd have no way to tell. How well did the re-recordings match the originals? If you spun her records over and over for a summer of your youth, I think you would notice any deviations from the original. Even the digitally-created audio would sound a little different unless she just outright copied the original files and used the exact same software. Anything recorded on real instruments would be even more impossible to match, with the absolute best session musicians playing the exact same instruments and using the exact same sound equipment you could get close but it would be noticeable.
I bought an iTunes version of Belle and Sebastian's classic LP "If You're Feeling Sinister" a few years ago and was bothered to distraction by the very slightly different mixing of the horn parts in "Stars of Track and Field", so maybe I'm an outlier, but if I've heard a song a lot it's like the uncanny valley effect to hear a mix that is only slightly different. I had to go back and buy the original CD off eBay, after checking catalog numbers to make sure it was the original pressing.
The re-recordings are intentionally very close to the originals, but they aren't identical. I would say that the main differences are that Swift's voice is more mature (particularly in comparison to the early albums) and the mixing is slightly better. There are a few tracks with very obvious changes, e.g. "Girl at Home" has different instrumentation; the lyric "She's better known for the things she does on the mattress" is cut from "Better than Revenge".
Taylor's music is deliberately crafted to be easy to sing along with, and that would be jarring if there were many changes. So most of the time, she stuck to the originals as closely as possible. In fact, for me it's sometimes hard to notice the differences.
But then I only started getting into Taylor in 2015, so perhaps someone who grew up with her music would think differently.
Thanks, this is really helpful.
I'm still confused by your last paragraph, though. What is the internal experience of being a Taylor Swift fan? I'm sure she's a very good singer and songwriter, but are people attached to her? Isn't her personality "mean girl who snipes at her ex-boyfriends"? Is this psychologically compelling enough to create an empire?
Swift is pretty careful not to be a Mean Girl in the sense that her potshots are limited to boyfriends, and her target market isn’t straight men.
For her audience she serves as a main character into which they can project themselves.
This is like the 5th Swiftie testament I've come across, and they all sound like an infomercial trying to sell me Taylor Swift in 7 easy instalments, or a letter asking for Taylor Swifts being Taylor Swift grant to be extended .
And the other things I have seen them doing: Get products offensive to their sensibility cancelled in an especially obnoxious way (dude, there's like 30 million of us. Do you have any idea what we could do without even trying hard? Take it down.), and wierd uncanny stuff like blank out on her concerts, and just post literal cheerleader chants when her name is mentioned.
I find her face and her name ugly and graceless. She released an anti-patriarchy song lamenting all the possibilities closed to her for being a woman. The messiah thing she must be cultivating on some level, and I don't like that either.
"Of all the options for Person of the Year, Taylor Swift is by far the most likable. That's why she's the overwhelming favorite in the reader's choice market: https://manifold.markets/Joshua/will-taylor-swift-be-the-time-magaz.
I hate how she doesn't even look any different to when she was first famous, at least Madonna pinned down distinct time eras every time she rose, Taylor Swift's annus mirabilis is just the last word in our culture being irrevocably stuck. Given her outsize influence, I'm going to unfairly blame her directly for this. This is what she wants. It's what she's wanted all along.
"At a time when the news is full of doom and gloom, and the country is more polarized than ever, Taylor is the one major person who is nearly impossible to hate. And that makes her the Person of the Year that we need right now."
Me. I have decided I hate Taylor Swift. What's more I hate every single thing I know about her. So I will hate her enough for everyone.
And she never answered this letter:
https://www.honest-broker.com/p/an-open-letter-to-taylor-swift
I asked someone to explain why they liked her to me and they did, so I don't fault them for trying to sell her.
I was writing a case for Taylor Swift as Person of the Year. I think she deserved it, but from what I saw here and on Manifold it seemed like most people didn't understand why she was even in the running. I wasn't trying to write an unbiased review.
In fact, we might actually agree on more than you think.
- I agree that her fandom is pretty toxic at times. That's not Taylor's fault, per se, but she sometimes takes advantage of it. For example, I felt like she should have told her fans to stop sending death threats to Scooter Braun. But it gave her leverage, so she said nothing.
- I assume by her "anti-patriarchy song" you're referring to The Man. I think she's probably right that she is scrutinized in ways that male singers generally aren't. But I find it a little cringy how hard she leans into the idea that being a woman is a disadvantage, while at the same time shrewdly using her identity as a woman to further her career.
- And yes, being famous is exactly what she wanted all along. In some of her songs, she's very self-aware about this. For example, take this line from Mastermind:
"No one wanted to play with me as a little kid
So I've been scheming like a criminal ever since
To make them love me and make it seem effortless."
You have to read between the lines sometimes to tell which of Taylor's songs to take seriously and which ones are a joke (e.g., Blank Space). But I'm confident that this one is absolutely serious - she's obsessed with fame to an unhealthy degree, and she knows it, but she can't stop.
So it's actually fine with me if you hate her. I don't think she's a saint, but I do think you're missing out.
> For example, I felt like she should have told her fans to stop sending death threats to Scooter Braun. But it gave her leverage, so she said nothing.
I am not sure if she counted this as a leverage for present or as a revenge for past (not that either is commendable, but both are somewhat understandable).
Swifites can be pretty unhinged. But in this particular comment thread, you're the one coming off that way.
Lmao this is an amazing comment
Taylor definitely isn't considered a mean girl by her fans. I asked my wife, and she called her a nerd. She was a band geek in high school that no one liked, and now she's an amazingly successful business woman who has upset industries and boosted the economy in presumably big ways.
"I'm sure she's a very good singer and songwriter"
This is a bit of a tangent, but I think to understand Swift's success, you need to appreciate her showmanship. I would say the order is songwriter > showman > singer. This leads to some confused people asking why she's so successful when her singing is no better than many others.
If you want to understand her success, you really ought to see her in concert. There are many Swifties, and no doubt they have a wide range of experiences, but I would say an important element for many is having shared in this extraordinary experience. Swift has a remarkable ability to create a feeling of intimacy with 50,000 people.
Unfortunately, tickets to her concerts are notoriously difficult to obtain. Fortunately, it's her birthday in a week, and the Eras film will be streaming on various platforms (e.g. Apple, Amazon, YouTube). I should warn you though: it's very long, even longer than the theatrical version.
That's a great question. I can see how you would get "mean girl who snipes at her ex-boyfriends" from certain media portrayals, but she's generally not mean-spirited.
I think what makes being a Swiftie unique is that Taylor is in love with her fans at least as much as they are with her. In fact, I suspect that's the reason many of her actual relationships haven't worked out - no guy will ever be first in her life, because her fans will always be more important.
For example, she's always made herself as accessible as possible. Her mom used to locate her superfans at her concerts and invites them to come meet her (though I'm not sure if that still happened with the Eras Tour).
After 1989 was released, she made this video of sending Christmas presents to her fans: https://www.youtube.com/watch?v=j3yyF31jbKo. And they weren't generic gifts - she stalked all of them on social media long enough to pick out a unique and special gift for each person.
And she constantly shares intimate details about her relationships through coded messages in her album notes. Taylor wants all her fans to know who each song is really about, even if she doesn't directly name the guy in the lyrics.
With that context, it's easy to understand why she and Joe Alwyn broke up:
https://people.com/music/why-taylor-swift-joe-alwyn-broke-up-after-six-years-exclusive/.
One of the best quotes from their anonymous source is, "[Taylor and Joe] had plenty in common and fell in love in a safe bubble while she was retreating from the world during Reputation... Then the pandemic hit, and they were locked down together and able to continue growing their relationship in this insulated way. But he didn't really 'know' her yet outside of that bubble."
And also: "Joe has struggled with Taylor's level of fame and the attention from the public."
That's obviously the way Taylor wanted people to see things. But reading between the lines, I think the real issue was that he wanted to remain first in her life, and she cared more about preparing for the Eras Tour. (Notice how she never seemed that disappointed by their breakup - she got what she wanted.)
So to sum up, the internal experience of being a Taylor Swift superfan is that the most powerful and most socially intelligent person in the world loves you deeply, and is constantly trying her best to show it. It's not so far off from the way Christians see Jesus.
And to be clear, I don't count myself as a superfan. I'm a "huge Swiftie" compared to most people here because I can sing all the words to most of her songs, but I wouldn't call myself that in a group of Swifties, because I don't worship her the way many of them do.
So she's Elvis?
Elvis was before my time, but it could be similar.
Maybe in 50 years people will go to Vegas to be married by Taylor Swift impersonators singing Love Story. That sounds way better than being married by Elvis. 😂
O.M.G.
One more item about how/why Taylor may not come across as a 'mean girl' is that she shows some awareness that the problems aren't all her boyfriends fault.
Some concrete examples from her songs:
*) The lyrics that open "Back to December" (and the entire song, really): https://www.kkbox.com/tw/en/song/9_tw43keL75jLQd0A9
*) The Blank Space lyrics. Again, the whole song, but specifically this:
Screaming, crying, perfect storms
I can make all the tables turn
Rose garden filled with thorns
Keep you second guessing like
"Oh, my God, who is she?"
I get drunk on jealousy
But you'll come back each time you leave
'Cause, darling, I'm a nightmare dressed like a daydream
*) The lyrics to 'Anti-hero' include this:
It's me, hi, I'm the problem, it's me
So Taylor does write songs about ex-boyfriends. She also writes songs where it is clear she is contributing the the relationships failing.
There is more, but that's a part of it. Her fans don't classify her as a 'mean girl' at all.
If you haven't seen her SNL monlogue it is worth watching. It's short.
I agree with you, but I think Blank Space is a poor example. Taylor has said that it's a parody of how she saw herself being portrayed in the media - it's not an actual confession.
The best case for Taylor as times person of the year is that it gets the Time's magazine the most attention.
Taylor is the anti-mean girls. The "you will always have friend and community while maintaining your unique specialness" product. They maintain an image, communicate the messaging people want to hear, and sell the product. Their songs are intentionally lower skill for singing along to take this even further. She occupies a niche and is not a reproducable approach to succeeding in the industry.
I don't think the Times is dumb enough to actually think Taylor is the person of the year for her own financial success (mega tour wow, omg congrats you made a movie too, AND you big dicked the big dicks!?!?). Times is smart enough to know that the swifities will bandwagon the decision and give Times tons of free media. Other people will question it with good reason and that will incite drama, more convo, more free publicity.
I wish I could send free psychadellics to every swiftie in america
> Why is Taylor Swift so high? I understand she’s a very famous pop star, but hasn’t she been an equally famous pop star every one of the past ten years?
I don't know why, but anecdotally, I've been hearing way more about Taylor Swift in the past 6 months than I did in several years before that. Seems she had some sort of resurgence in popularity.
Regarding long-term markets, I had a thought on early resolutions. Specifically, I created a market that asks if a large construction project will finish in time. It will resolve as NO as soon as a delay is announced. This is good because it reduces the expected period of capital lock-up. This is also bad because it creates different incentives for different participants: If you expect major delays, you also expect a shorter lock-up. If you expect a punctual finish, you expect a longer lock-up.
I'm basically resolving the market early based on a (somewhat authoritative) third-party estimate. This is problematic in itself, but seems fine to me because the project management organisation is strongly incentivised to not falsely announce delays.
I was wondering if anyone had thoughts on this approach.
I will also shamelessly self-promote my Manifold market: https://manifold.markets/ThomasTwenhoven/will-the-fehmarn-belt-tunnel-open-i
"The best realistic medium-term outcome I can imagine for the people of Gaza is as something like a West Bank without settlements and roadblocks"
I don't think this is even remotely realistic. Israel has worked_ hard_ to cleanse the good areas for colonization and settlements - there is absolutely no way they would ever return all of that, not even if a credible peace was offered (and no-one can offer that). That ship sailed long ago - it could _possibly_ have happened in the early 2000s, but not now.
That has nothing to do with Gaza, which hasn't had settlements since 2005.
Thanks - re-reading I _think_ what Scott is saying isn't fixing the West Bank, but having Gaza in the same kind of statelet. That's probably doable, at least after Israel strips away a chunk along the border to get a wider (and obviously necessary) security zone.
So mostly status quo ante bellum (apart from, ideally, no Hamas for a while), alongside _some_ territorial losses. The blockade will obviously remain, as will even more draconian entry checks into Israel.
I don't know much about the Middle East - why will the blockade obviously remain? I thought the blockade was in place because Hamas kept converting imports into rockets. If Israel has enough control to keep the terrorism down, couldn't they relax the blockade?
Israel _won't_ be able to control terrorism. They might (possibly) be able to ensure the terrorists aren't actually in charge, but terrorist groups are going to remain and pop up.
Two long-shot possibilities that I mentioned on reddit in https://www.reddit.com/r/TrueUnpopularOpinion/comments/17xmm1x/comment/k9q1gfq/
Two possible thoughts, one using technology that I'm not sure of:
1) If lie detectors work today, have each Gaza or West Bank resident swear that they will never attempt to kill Israelis while on a lie detector, and allow the peaceful ones to stay.
Alternatively:
2) Have a two-state solution, but the Palestinian state is not self-governing, but is rather administered by the People's Republic of China, analogously to Xinjiang. Admittedly the Uighurs only killed ~100 Han Chinese in their attacks, a dozen times less severe than 10/7/2023, but the PRC seems to have the will and the technology to manage Islamic minorities safely at this point. I think the PRC could prevent a Hamas2.
Lie detectors detect lies more accurately than chance (which is better than most humans), but are mostly useful as security theater. They're not nearly accurate enough to do 1 particularly well.
Two things:
(1) I am totally uninterested in talking with, using, or playing with ChatGPT or any of the competing chatbots. "But you can do art and write essays and this, that and the other!"
Yeah. I find I don't care. I tried one model that allows you to just start online, and I found myself sitting there with a blank input prompt box, because there was nothing I wanted to ask it, no conversation I wanted to begin, not even any "can I make it do no-no things?" because I don't care what a machine spits out as 'Hi I'm your real liddle fwend who wanna talk wid u about stuff just like human person!'
So I'm approaching this whole giant blob from the point of view of "uninterested idiot"
(2) Given all the chit-chat about OpenAI, here's what Microsoft are doing in the field. This isn't theory, this is "selling a product to businesses". I think it might be interesting to get the view from the outside, as to what the actual implementation of AGI is visualised as (not the theoretical paperclipper/fairy godmother stuff):
https://emails.microsoft.com/dc/K2nLtwtctcW29fgYHxJx83uewKopb_trQgTHLElmnWB9QXBiJmVo0LJiMhILVKH4QFl4GlquDHOUOwkIJxSHvkFr23HfHwEsECLLexVuE8fj03PGrSo07AxezE-XLL0c/MTU3LUdRRS0zODIAAAGP2GptF7TJAshmzRcBEUUQ-sLidWqdab7nU6pLZWnLYy0igg3wAyEyintvu1TaPQtC-2_Y1NM=
That mess above brings you to the PowerPoint Presentation from the November "Microsoft Discovery Day: Building AI-Powered Apps". It amuses me, a bear of very little brain, that nestled in among all their salesmanship about how ground-breaking the AI is, there is still an AI-generated piece of art which can't understand how to spell correctly "integration" or "recognition" in "AI integration" and "Voice recognition" text on image.
"Microsoft and OpenAI partnership
OpenAI: Ensure that artificial general intelligence (AGI) benefits humanity
Microsoft: Empower every person and organization on the planet to achieve more
Azure OpenAI Service :
GPT-4 & GPT-3.5-Turbo Text, Chat & Code
DALL·E 2 Images
Whisper Transcription & Translation"
Very nice, how do they see this translating into practice?
"Introduction to Top Use Cases
(1) Business Problem
Productivity is lagging
Business Needs
Increase Productivity
Solutions
• Conversational Search/Knowledge Insights
• Code Generation and Documentation
• Trend Forecasting
• Report Summarization & Generation
(2) Business Problem
Need for process Automation
Business Needs
Automate Processes
Solutions
• Document Processing
• Workflow Management
• Fraud Detection
• Supply Chain Optimization
(3) Business Problem
Degraded Customer Experience
Business Needs
Improve Customer Experience
Solutions
• Intelligent Contact Center
• Agent/Employee Assistance
• Virtual Assistance
• Call Analytics
• Call Summarization
(4) Business Problem
Creating Content is Time Consuming
Business Needs
Build Creative Content
Solutions
• Marketing/Sales Content Generation
• Personalized Content Generation
• Product Design & Development
• Digital Art
What can Generative AI Do?
Generate New Revenue Streams
Deliver Differentiated Customer Experiences
Modernize Internal Processes"
In future, Scott can be freed from the time-consuming burden of "build creative content" by outsourcing it all to AI. How nice to look forward to this!
So the AI future in the workplace is not going to be increased leisure and money, it will be "you WILL be more productive! you WILL increase velocity! if you still have a job, that is, and have not been replaced by the machine altogether". I honestly think *that*, rather than "unaligned AI will find a way to 3-D print neurotoxins to be delivered globally by nanobots to wipe out humanity" is the real problem we will be facing and should be discussing.
"The only error is that it recommended a deli counter as if it were a sit-down restaurant."
Be the change you want to see in the world! Enable sit-down dining at the deli counter now! Grab a few rolls from the bakery section, a bottle of the beverage of your choice from the wines and spirits, and there you go!
...bring a folding chair with you ! :-)
Commandeer one of those little stands/stalls they have for product demonstrations, there's always one of those in a supermarket sometime.
Plus you get to eat the free samples, it's a win all round!
I can't understand you not wanting/needing to experiment with it. As far as not knowing what questions to ask, you can ask it the same things you would ask Google, or any other search engine. It isn't about conversation, but about finding answers and doing language busy-work.
Of course, it is about as accurate as any source found on the internet, and you must bear that in mind with any output.
But it can produce average quality written material, without typos, or standard messages. Here are some things you can ask it with good results (after critically reviewing and revising the output):
- Give me a job description for a bookkeeper position that does both accounts payable and accounts receivable work, among other things.
- Why does ice take up more volume than water?
- Write a two-page essay about why dumb phones can be better than smart phones.
I think, looking at my own reaction, it's because I don't see it as anything other than a machine.
So I'm not interested in whatever it outputs, since I know it's been formatted to pretend to be a 'personality' and reply to me that way.
If I want to look something up, I'll go straight to Google, and there isn't any "language busy-work" I need it to do; I'm long past having to write essays for homework, thank God, my workplace has standard job descriptions in place if it is hiring on new staff, and for work it's generally short emails in response to boss or colleagues.
Now if I needed to churn out blah blah meaningless reports (like a previous job) I might be happy to hand that over to it, but right now I don't, so there's nothing there I care about using.
I'm very surprised by this myself, since I would have imagined I'd *want* to experiment with it, but I think part of the hype has put me off and the rest is "yeah - it's a box, it's not thinking or understanding, why do I care? I don't".
Of course it's a machine. Those who think it can actually think or has consciousness I believe are misguided. But you probably have other things you may want help with, or which could spur your own creativity.
What about writing a thank-you note to someone you don't really know? What is an appropriate congratulations message to an acquaintance? I'm sure you get stumped on various kinds of writing like everyone else does, but the machine is never stumped.
"What about writing a thank-you note to someone you don't really know? What is an appropriate congratulations message to an acquaintance?"
"Thank-you notes to people I don't know" are part of the bread-and-butter of my current job, and I have it reduced down to a formula which I think nicely balances oily flattery with not seeming too obvious about it 😁 By the time I've prompted the ChatBot, removed the Americanisms, and read over and corrected the thing, I'd have the letter written.
Too much information about my personal life, but I don't write letters, notes, emails, etc. to acquaintances and friends 'cos I don't got none. All that writing is on the job, any 'personal' stuff is via social media and I curate that kind of thing I say along strict parameters. EDIT: There are also plenty of old-style letter-writing guides and new model letter examples online, so while I could find one of these AIs to do it for me, it's just too much bother. "Hey, Chatty, here's the person, here's the sitch, here's what I want you to say, make it informal and breezy, yadda yadda yadda", get the output, keep going through it, finally get a finished product to bang off - I could, but that sounds like climbing onto a pair of stilts to walk up a ladder in order to reach the top shelf of the kitchen cupboard, you know? Quicker and easier for me to go on my tippy-toes.
When I'm writing for myself, as in fiction writing, I prefer to do it my own way. Having the bot churn out some pap that's tailored to "and you said you liked marmalade, rain on the sea, and violent torture of the woobie" to 'personalise' it to my requirements - no, thank you!
You can’t think of *any* busywork you’d ever want to outsource? LLMs are the written-language equivalent of a graphing calculator for math; they are good at doing boring repetitive stuff to save your time to work on the important items.
Even if you don’t have anything like that, surely you can understand why other people would find it useful. For example, I have no interest whatsoever in writing tedious marketing emails, but that could be a big part of my work that I can just make an AI do for me!
Oh, I can see the potential for automation, but right now what it can do is not what I need done, or I don't have the resources to link up an LLM to the things it needs to scan in hand-written documents, etc. and turn those into intelligible data.
The impression I get, and I am probably wrong, is that I have to babysit the thing while it churns out response to the prompt, so it doesn't save me time to get on with doing other things. As it stands, I certainly would *not* trust "get LLM to write that reply email and send it off to Revenue or the Department of Education" without double-checking it first.
> I think, looking at my own reaction, it's because I don't see it as anything other than a machine.
That's not the real problem; the real problem is that it's a *really stupid* machine. ChatGPT is miles ahead of ELIZA in terms of grammar and syntax, but not very far ahead at all in terms of actually delivering useful information. As you'd said, Google can deliver the same data, in a more reliable format.
That said though, I think that image generators *do* have a useful purpose: in our RPG gaming group, we use them regularly to come up with character pogs. Would we still use them if at least some of us had artistic talent ? Probably not, but none of us have such talents, so...
ChatGPT has occasionally done better than Googling stuff myself, because when you're trying to learn something new it can be useful to aggregate the result of many Google searches into a single summary.
For instance, I was looking for a knife sharpener recently, and Googling it turned up a ridiculous number of possible options - do you use electric or manual, ceramic or tungsten carbide or diamond, the kind with spinning wheels or the kind with v-shaped grooves, how much material does it take off, what angle does it sharpen at, do I listen to the people on Reddit saying I need to learn to use a whetstone, etc. So after Googling around for a while, I went to ChatGPT and said "Tell me about the different types of knife sharpener and what the pros and cons are" and got a nice summary of everything.
IMHO, ChatGPT exists to summarize the internet for people, in essay form, and perhaps 99% of the internet is digitized mass communication. It's a real time, automated version of Wikipedia, without the source notes (well, without reliable ones). The way to make the most productive use of it is to carefully craft the prompts to correspond to sources you have already checked. I have no objective evidence, but I'm convinced that it makes the most money for users by being a marketing tool.
If that sort of thing does not interest one, then the use cases go way down.
"Here are some things you can ask it with good results (after critically reviewing and revising the output)"
That parenthetical is doing a lot of heavy lifting, and in a way that negates a lot of the value of LLMs at least for applications I care about. ChatGPT's answers *cannot* be trusted for anything important. We had the example right here of ChatGPT all but advising SpaceX to do, in the name of increased "safety", something that would turn every Starship into the equivalent of a tactical nuclear weapon with a hair trigger.
If the question is simple, and if I am confident that I know the subject, then I can proofread and revise very quickly. But for simple questions in my area of expertise, I can come up with my own answer very quickly, perhaps not quite as quickly as prompting and proofreading and revising ChatGPT, but pretty close and with less possibility of an embarrassing goof. If the question is not simple, or not in my area of expertise, the proofreading part approaches just researching and solving the problem myself.
I can see ChatGPT and its kin as useful tools for generating pointers to answers, but I've already got search engines and wikis for that.
For generating probably-accurate trivia of no particular importance, sure, that's going to be fast and easy, but why should I care?
" We had the example right here of ChatGPT all but advising SpaceX to do, in the name of increased "safety", something that would turn every Starship into the equivalent of a tactical nuclear weapon with a hair trigger." Yup! That was dramatic.
Once upon a time, we were considering Project Orion as a serious proposal for efficient starship propulsion. People can come up with dumb ideas, too.
Project Orion essentially was using small nuclear bombs to lift large masses by action-reaction, exploding them just behind the starship.
Yes, I've read of it. I suspect that a large chunk of LLMs' current problems come down to (to anthropomorphize) they being insufficiently self-critical. And one can be wrong in either direction, being too self-critical or insufficiently self-critical. And it isn't so much that Project Orion was a _bad_ idea (the expected specific impulse would have been much better than chemical fuels) as that it got nixed by the limited test ban treaty and, by now, the combination of improved chemical rockets and (for low absolute thrust, high total delta-v) the combination of ion or plasma thrusters plus nuclear reactors made Orion moot.
The people proposing Project Orion were not factually wrong about what would happen if an Orion were built and launched. They recognized both the costs and the benefits, and considered the benefits worth the costs. You may disagree, but that doesn't make them or their ideas dumb.
ChatGPT, is just plain dumb. It knows a lot, but it's too dumb to realize that it doesn't know everything, and it has great difficulty learning new things.
No dispute that ChatGPT is dumb.
But did they really consider the fallout from the explosions? The trail of it through space? Do they also explode nuclear bombs to slow down? Maybe it could be considered as a plan of last resort to escape forever from a doomed planet, but it's hard to imagine actually using explosion after nuclear explosion to power a starship.
I'm mostly probing to see when the LLMs stop hallucinating.
I just tried ChatGPT again, this time asking for the elements, sorted in order of increasing density, and it didn't sort them correctly. (Ok, it was the free version, still, this isn't new or difficult information)
Dont ask it for data. Ask it to summarize data you give it (you point it to where you want it to go--in your case, something like summarize this paper (contains the table of periodic elements)).
It's a copy writer, not a researcher.
Well, I was hoping for better from even "a blurry jpeg of the web". A good chunk of the potential value of an LLM trained on a vast quantity of information is to have it pick out data from a hundred places _without_ the user having to explicitly specify those places. If it can't do that, so be it, but then the technology is significantly less valuable :-(
The problem here is that the vast quantity of information that the LLM is trained on, the internet, does not itself consist of reliable answers. In fact a fair chunk of it is misinformation. So long as we are the ones training it, it isn't ever going to be more accurate than we are.
Roughly speaking, I agree with you.
More precisely, where there is general agreement by writers of the training set on some question, if the agreed answer is wrong, the LLM will get it wrong as well.
As LLMs that I've been able to play with recently stand, the LLMs behave substantially worse than that. In the example that started this subthread, I asked an LLM (Bard, in this case) to list the elements in order of their densities (of their condensed matter phases). This isn't the sort of data where there is a lot of misinformation on the web. The right answers are pretty much in Wikipedia, and, as far as I know, there aren't widely promoted wrong answers. In fact, the _numbers_ that Bard came back with looked about right. Weirdly enough, it got the _sorting_ wrong. So this is a different problem from misinformation on the internet.
In principle, an AI that uses the internet as a training set can do better than the raw error rate of information on the internet by looking for inconsistencies. If it sees five people make one deduction from a set of premises, but they all have logical errors in their arguments while a sixth person starts from the same premises but makes logically valid deductions and arrives at a different conclusion, a sufficiently sophisticated training process could theoretically notice this and arrive at better conclusions than raw statistics would yield.
I'm still waiting for worse blunders to be fixed. :-(
(1) For me it is simple; ChatGPT replaced a part of search engine usage. If I am looking for a specific web *page* (for example a YouTube video), or I want to read a *document* in its entirety, I use Google. If I am looking for an *answer* to a question, I ask ChatGPT instead.
Ultimately, using a search engine to find answers is very ineffective, because a search engine only finds *keywords*, and those can appear in unrelated context, and can be gamed. In best case, you still have to click on the page, click through the pseudo-GDPR "consent" forms, scroll down, and find the answer in text. ChatGPT instead answers your question directly, immediately, using the knowledge from many web pages; sometimes it even provides additional useful information.
It does not have to be super-intelligent to be smarter than Google. Some things are so SEO-spammed that it is virtually impossible to find the answer by googling. Try e.g. "Japanese words starting with v"; if you find an actual answer on the first page of Google results, I am jealous. ChatGPT just gives you a list.
For a software developer, ChatGPT can also done some "manual" work for you. Something that a junior developer could get done in an hour, except that ChatGPT does it in a second.
(2) Sure, "2x productivity" has always meant that you will have to produce 2x more in a unit of time. Not that you will have 2x higher salary, nor that you could get a part-time job now. For an employee, the only difference is having to learn how to use a new tool (preferably before the job interview, during your free time).
The problem with ChatGPT as a replacement for search is that sometimes it thinks it knows the answer but it's confidently wrong. So if it's important you have to look it up anyway. If you use the web-enabled version it's even worse because it just gives you a summary of the top few bing results even if they suck. I agree that it can be good for certain questions that have no good search keyword, or where no high ranked page has an actual answer.
I think ChatGPT (particularly the speech version) is useful for satisfying casual curiosity. If I'm reading about, say, the constitutional referendum in Chad, and I'm not familiar with the background, I can just ask it, get an answer, and continue reading.
Did you know that purple used to use straw, also called "thresh", instead of carpet? It was spread in a thin layer on the floor. To keep it from spilling outside, they put a bar at the bottom of the door. This was the "thresh-hold".
Except that isn't actually true. But I bet it's easy to find that story on the internet, in lots of places. ChatGPT might even give you that story.
Don't believe what you read on the internet until you can confirm it. ChatGPT isn't more trustworthy just because it's a machine.
"Sure, "2x productivity" has always meant that you will have to produce 2x more in a unit of time."
Which is how new tech has always been implemented, from the invention of the typewriter onwards. But it gets *sold* to us, the cogs in the machines of commerce, as "this will lighten your workload and reduce your work day". Nope. Free time remaining now the new wotsit has sped up production and whizzed through what used to take four hours to complete? Don't imagine you can clock off early, now you need to do 2x the work in the same time.
> Don't imagine you can clock off early, now you need to do 2x the work in the same time.
This is a logical consequence of the global elite's century+ run of spectacular success at making sure that no appreciable increase of abundance ever happens to the actual necessities of life for which "rat race" inmates (read: you, I, and everyone you're ever likely to meet) "race to the bottom" in competition: comfortable housing; freedom from criminal predation; such food as existed 300y ago (i.e. not Velveeta, "McFood", antibiotic-laden meat, palm oil "chocolate", etc) ; health (which is not 100% equivalent to "access to medical treatment", as Americans were led to think) ; insulation from financial volatility (i.e. "sorry, you don't get to have a roof over your head now because someone-on-the-other-coast-did-something-with-stocks" etc.)
See also Moly's "involution" :
"Involution is basically…a race to the bottom, stressful competition with no winner, basically. The best explanation I’ve seen of it is that if everyone was sitting in a theatre watching a movie, and all of a sudden, to get a better view, somebody stands up to watch. So the guy behind him has to stand up, and the guy behind him has to stand up. And soon enough, everyone in the theatre is standing while watching the movie. Everyone is less comfortable, they’re being made to work harder, and yet the movie is still the same movie. They didn’t get anything extra out of it at all." ( https://weibo.substack.com/p/110323-faq-answered )
... or Scott's "Moloch" piece.
"Involution"/"Moloch" is not in fact an inescapable natural phenomenon. It is engineered -- by the people who prevent the proverbial theater audience members from whacking the first stander-up on the head with a heavy blunt object. Which people? Our rulers. In every case.
I hope you double-checked that list because I'm pretty sure Japanese doesn't have a "v" sound except sometimes in loan words.
Yes; I wanted to find those loanwords.
But most pages for "Japanese words starting with v" -- heck, most pages with TITLE "Japanese words starting with v" -- actually contain *English* words starting with v. There are tons of such pages, probably for Scrabble players or something like that.
ChatGPT explained that only the loanwords starts with v, and then gave me some examples.
I'll reiterate a "prediction" I've been offering recently: We'll arrive at a substantial understanding of the internal structure and operations of LLMs BEFORE (the mythical) AGI is attained. Moreover, once that happens, the need for belief in AGI will weaken and the idea will disappear, as did the idea of phlogiston once upon a time. What will replace AGI? I surely don't know, specifically. But in general, a richer and most robust understand of the mechanisms of both artificial and natural intelligence.
On natural intelligence: Principles and Development of Natural Intelligence (5 of them): https://www.academia.edu/235116/Principles_and_Development_of_Natural_Intelligence
comparing phlogiston and AGI is weird to me because AGI doesn't actually propose any specific mechanism. its like comparing phlogiston and combustion.
Even better.
The concept of "combustion" didn't vanish from our vocabulary when we discovered that phlogiston wasn't a real thing, because "things burn" is still a real thing and "combustion" describes that, regardless of if the mechanism is phlogiston or chemical energy.
Similarly, even if it turns out that we're wrong about the nature of consciousness and how LLMs "think," the phrase "general intelligence" isn't going to vanish from our vocabulary, because "the sort of thinking that humans can do and current-generation LLMs can't" is a real thing, and you need a word for that no matter what the underlying cause is. It describes a capability, not a mechanism.
What's going to happen is that we're going to come up with mechanisms to account for capabilities. In the process, we may have to revise our sense of what capabilities are possible. Just because we have a name for a capability doesn't mean that it's a real thing. If we don't know how it works, then there's a sense in which we don't know what it is.
Yan LeCun keeps making the point that, no, we don't have general intelligence. We can do a lot of things, but not everything. Moreover, each individual has specific capabilities.
So, what cognitive capabilities are humans lacking that we can't be called "general intelligence"? Because it sure seems like our intelligence can be applied to a broad variety of fields. We don't have a specific lobe of our brain that enabled us to invent quantum physics, we have a general process of understanding the world which works for every field of science.
If an AI is able to do everything that an average 100 IQ human can do, without needing to be retrained between tasks, I think most people would call that "general intelligence," regardless of if Yan LeCun thinks there are still secret cognitive tasks to discover which neither humans nor machines can do.
Yup. I view "able to do everything that an average 100 IQ human can do" as a goalpost (albeit one with some haziness). How much detailed understanding of mechanisms (beyond training neural nets) is needed to reach that goal is an open question.
not even mentioning how most people include humans when they talk about general intelligences anyways
Where's the 'intelligence' reside? In the 'raw' human brain, or in the brain that's been specialized of this or that subject? Is your brain specialized for every subject under the sun? Is anyone's? What are you imagining for the AI of the future? One giant humongous 'intelligence' of 10,000 AIs communicating with one another, each of them variously specialized?
This isn't a serious intellectual discussion. It's bar-room chitchat. That's fine in its way, but...
Just because we can string a bunch of words together in a coherent way doesn't mean that they refer to something that's real, if not in the past or present, then possibly the future. LLMs do it all the time, but sometimes they're "hallucinating," "confabulating" is a better term, BTW. And that's what talk of AGI etc. is, confabulation, a collective confabulation.
Your prediction isn't relevant unless LLMs scale all the way to AGI/ASI.
So with that in mind I predict that your prediction will be wrong in precisely the scenario where it matters. Pretty decent chance of you being correct in the case where it turns out to be irrelevant though.
Do you think the successful AGI/ASI scenario will happen With or Without machine learning? We aren't going to get there unless we understand what's going on inside ML models. Basically, we're dealing with associative memory. Details will vary, etc. But the principles are the say. We just have to figure out what they are.
I revise my prediction to a general understanding of the mechanisms of ML models must come before AGI/ASI.
I don't know if you are right or wrong, but I can think of an alternative: We get AGI when we figure out what conceptual self-awareness really is.
InB4 someone says SA is simple and we already understand it: obligatory reference to the XKCD cartoon of the physicist wondering why other fields need whole journals for them.
"Just model it as <Simple Object>, and then add some secondary terms to account for <Complications I just thought of>. Easy, right?
Doesn't AGI just mean "AI that seems about as smart as a human"? How do you debunk that? What would that even mean?
Right, just what does that mean ("AI that seems about as smart as a human")? How are you going to specify it? An IQ range? Scores on college entrance exams? Maybe you just talk with it and arrive at an intuitive judgement.
Refresh yourself on the initial remarks Steven Pinker made in his debate with Scott Aaronson in 2022: https://scottaaronson.blog/?p=6524
Well, at the very least (and I am very much *Not* proposing that this is all we would need to do, just this as a first tentative step) we would need to create the functional equivalent of the networks of associated schema and prototypes that the Long Term Memory uses to store the idea of every life experience we have had that meant something to us.
To set the record straight: I once gave a ~2% probability for the classic AGI-doom paperclip-maximizer-like scenario. I have a much higher probability for an existential catastrophe in which AI is causally involved in one way or another — there are many possible existential catastrophes (nuclear war, pandemics, runaway climate change…), and many bad people who would cause or fail to prevent them, and I expect AI will soon be involved in just about everything people do! But making a firm prediction would require hashing out what it means for AI to play a “critical causal role” in the catastrophe — for example, did Facebook play a “critical causal role” in Trump’s victory in 2016? I’d say it’s still not obvious, but in any case, Facebook was far from the only factor.
I would argue that there is pretty solid journalistic level evidence that it was a contributing factor (and not by accident either--it happened on purpose). Given how close the election was...
https://en.wikipedia.org/wiki/Facebook%E2%80%93Cambridge_Analytica_data_scandal
Also, the GREENS took some votes from Hillary Clinton. And guess who had lunch with their candidate? Vladimir Putin. Misha Frier: "Jill Stein launched the 2024 White House bid as a Green Party candidate. Is she getting any support from Mother Russia again?
Back in 2015, after the Annexation of Crimea and the launch of the war in Donbas, Jill Stein was at a gala dinner in Moscow to celebrate the 10th birthday of the Russian TV propaganda network Russia Today. She was there at the invitation of Vladimir Putin and sat behind the same table with him and two former KGB agents, Gromov (chief of propaganda) and Ivanov (Putin's chief of staff), at the gala dinner.
In 2016, Stein's presidential campaign was heavily promoted by RT. I mean what better person to ask for endorsement of an environmentalist candidate that a dictator who bases his power on oil and gas sales to the West?
Stein didn't win, but she contributed to Donald Trump winning the election that year. Her vote totals in the crucial states of Pennsylvania, Wisconsin and Michigan arguably denied Hillary Clinton an Electoral College victory." - Also CNN https://edition.cnn.com/2016/11/10/politics/gary-johnson-jill-stein-spoiler/index.html
No AI, just ol` KGB-I.
To be clear, your 2% risk is from being paperclipped (inner alignment failure + fast local takeoff?) and the probability above 2% comes from exacerbating non-AGI risks? Does that mean you don't have much room for non-paperclip AI catastrophe, like rapid value drift from substituting ourselves out of the economy or fast local takeoff from a badly outer-aligned AGI, etc?
No, all those other things go into the beyond-2% zone.
Why 2% Scott? Or rather, how did you get to the figure?
I find it a little bit funny that you give it basically the exact same probability as P=NP :
"I’d give it a 2 to 3% chance that P equals NP"
https://www.technologyreview.com/2021/10/27/1037123/p-np-theoretical-computer-science
> I thought Kalshi was trying to cultivate good will with the CFTC, and this seems pretty adversarial.
A couple of perspectives. First, cultivating goodwill can be (is?) adversarial. If nothing else, you're competing (sometimes with the CFTC) for the CFTC's attention.
Second, even if the parties are getting along, an adversarial process might be inevitable to break a logjam. When an agency reaches the limit of its statutory authority, its hands might be tied (especially if there's political or media scrutiny). In that case, you go to Congress for a new statute or you go to the courts for a new understanding of the statute. Either way, adversarial.
To be clear: I'm not an expert in CFTC, prediction markets, or regulatory law in general. But I am a lawyer at a state-gov agency. We work very closely with regulated entities and have a strong customer-service culture, but it's often clear that an entity has to turn to politicians or judges to get around limits on agency power.
I've seen that from the other side in a previous job; the regulations based on the relevant legislation can be interpreted in two ways; the wording is vague enough that either interpretation is plausible; and nobody wants to be the one to stick their head on the chopping block* and say "yes, make decisions based on interpretation 1/interpretation 2", all the way back to the head office in the government department, until there's been a law suit that results in a ruling that "interpretation X is the correct one". After that, you can't get into trouble, because "well the courts told us we had to do it this way".
*Because if you say "do it by interpretation 1", you are spending more money, and the Department will not be happy with you and you will get yelled at, you will get your bosses yelled at, and your bosses will yell at you about over-spending the budget
If you say "do it by interpretation 2", you don't spend the money but then the aggrieved client who was refused may well go to court to force the local authority to spend this money. This is good in that it does establish a precedent, but bad in that you will get yelled at and get your bosses yelled at for dragging the Department into court.
Taylor Swift, yet again beating out Beyoncé despite Beyoncé also going on the biggest tour of her career.
Taylor Swift had the most lucrative tour ever, the most lucrative music film premiere, and she has a football player boyfriend.
Could care less about Beyonce, but I was rooting for Elton John.
>> To my surprise, this not only hasn’t collapsed, but has attracted people outside the usual prediction market community:
These are fake profiles. Some with locations like "California City" and "Arizona City".
TOTALLY NORMAL NON-ROBOTIC PLACE NAMES.
<mild snark> (nominative determinism?)
Since it is _Manifold_ Love:
Does the manifold have a metric? Does the metric satisfy the triangle inequality? Does this interoperate with romantic triangles?
</mild snark>
Those sound just as fake as Oklahoma City and Kansas City. Not to mention Mexico City.
American town names are such that I didn't blink at "California City" as "obviously fake".
And turns out there really is a California City:
https://en.wikipedia.org/wiki/California_City,_California
"California City is a city located in northern Antelope Valley in Kern County, California, United States. It is 100 miles (160 km) north of the city of Los Angeles, and the population was 14,973 at the 2020 census. Covering 203.63 square miles (527.4 km2), California City has the third-largest land area of any city in the state of California, and is the largest city by land area in California that is not a county seat.
Much of the workforce of Edwards Air Force Base, which is located 18 miles (29 km) southeast of the city, is made up of city residents. Other major sources of employment include California City Correctional Center (California Department of Corrections and Rehabilitation); Mojave Air and Space Port and its flight test operations; and the Hyundai/Kia Proving Grounds located in the rural southwestern part of the city. California City has a park, a PGA golf course, and a municipal airport."
Now I want to know, is there a California City *not* in California?
And imagine the conversations you can have!
"So, what do you do?"
"Oh, I'm a city clerk"
"Great! Where?"
"City of California City" https://www.californiacity-ca.gov/CC/
"In... California?"
"Where else?"
Where else, indeed? 😁
Arizona City is also a real place: https://en.wikipedia.org/wiki/Arizona_City,_Arizona
As is California, Maryland: https://en.wikipedia.org/wiki/California,_Maryland
"So, where are you from originally?"
"California"
"Ah, right: a Marylander" 😁
I'll call your "California City" and raise you one of these:
https://maps.app.goo.gl/fcVVL6V3JcdYYeij7
"Climax, Michigan", anyone?
I'm entertaining the idea of trying manifold.love, but I have a physical preference I perceive is the opposite of what people would expect. It seems like that would be important information for anyone to make better bets, but I normally avoid putting that publicly on my profile in dating apps because because I'm *open* to women who don't match that preference and I don't want to burn that bridge with them (for them to know I have a preference they don't meet). How would it make sense to handle that on manifold.love? Maybe accept that this is the one dating app where I should risk ruling out all women in one category for the sake of better matches?
re: Kalshi
I think the details of Kalshi's proposal for election markets are very important. iirc, the original proposal was markets on the outcome of Congressional control with position limits of $25,000 -- seemed fair enough. But their most recent proposal allows ECPs (eligible contract participants) to invest up to $100 million, and I think this is just way too high a limit to start playing with, and I worry concerns about manipulation actually become quite plausible. I get why Kalshi did this, because the CFTC said that they didn't believe the contracts had significant economic value for hedging with a 25K limit (which oil company is going to stake 25K on an election market to hedge against Democratic control), but I think the better route is to start small, and see what problems arise, instead of just opening up election betting on such a massive scale all at once. I reckon it will probably not end up being a good idea to have such high position limits to start, if ever.
If the court rules on the narrow question of whether "election markets" are or are not "gaming", that could still give the CFTC enough leeway to regulate contracts for their manipulability (warning: not at all a legal expert). But if the court just totally guts the CFTC's ability to regulate election markets, I think this will be bad -- I think if you want markets to work well (not just in terms of accuracy but social legitimacy), they need to be regulated (imposing requirements like KYC, market surveillance, etc.).
"But their most recent proposal allows ECPs (eligible contract participants) to invest up to $100 million, and I think this is just way too high a limit to start playing with, and I worry concerns about manipulation actually become quite plausible."
Well how can that possibly be risky, it sure isn't like we've seen a magic money machine founded on Rationalist pet project lines with big sums of money floating around going very publicly "ka-boom!" in the press recently with attendant trials upon trials 🙄
I gotta say, the endorsements feature on manifold.love sure seems underused. :P
Scott: "An Israeli occupation would involve constant bloody resistance; I'm not convinced it would be any better for people on the ground than Hamas (though someone can try to convince me otherwise)"
Well, one possibly convincing argument is that we have a test case for what it means for Israel to occupy a chunk of Palestine with a couple million Palestinians in the 21st century, and it sure seems to be better than Hamas. Israel has occupied the West Bank continually for decades, with soldiers, police, and settlers. There has been resistance, and I suppose pedantically you could call it both continuous and bloody, but I'm guessing you were thinking of something more than that.
And while I suspect that both Israel and the non-Hamas Gazans would prefer that Gaza be occupied by UN peacekeepers or maybe even Egypt, but those aren't likely to happen. But I don't see any fundamental reason why an Israeli occupation of Gaza, particularly in the medium term, should be any more bloody than the West Bank is now.
On the one hand, at least in the short term the Israelis have given the Gazans cause to be really, really upset with them. On the other hand, the Gaza population is presently going through a filter that will select for not being eager to engage in bloody violent resistance to Israel. And there are no settlers in Gaza, nor are there likely to be, which eliminates a huge ongoing aggravation.
I just fixed the issue with the embeds. There was a difference in the embed graph code that was making it connect the points for each bet with straight lines. Which is wrong! We put every single point* so the chart should be stepped rather than linear interpolated. Usually this wouldn't be a significant visual difference but in this case nobody bet on the market for an entire month before it dropped.
*[skippable nerdy aside]: We used to literally put every point on the chart but since lk-99 this would be really slow on the most popular markets. And on a chart that is at most 700px wide people can't perceive the level of detail from tens of thousands of bets anyways. At the same time I HATE those smooth curvy trailing average charts that ui designers keep using because they don't understand that data should not be beautiful; data is bumpy, sharp, and real. It took some tinkering but the solution I settled on for big markets is to slice the x-axis into 500 evenly-sized buckets; take the min, max, and median in each bucket (or the last value in the previous bucket if this one is empty); and draw the chart with those points. I am pretty proud of the end result, I think it looks very close to what it would look like with full data. As long as you don't zoom in :p
Oh, also we start the calculation with no more than 50,000 bets on a market, semi-randomly sampled, because I was too lazy to paginate through all the bets and eh it works
Manifold.love is a neat idea. I'm all in favor of anything that comes close to bringing back 2005-2010-era OKCupid, and the matching questions are a big part of that.
That said, I'm dubious of the "make a Yes bet to send an intro" mechanism. It suggests some specific target user demographics: people who have more than a 50% chance of a first message turning into a first date. As far as I know, that niche contains most women and very few men.
Are there guys on the platform making confident Yes bets of any substantial size, or is everyone just putting in Ṁ10, or are people spending real money to buy Ṁ to make bigger mostly-losing bets on intros, or... something else?
I don't think a Yes would have been a good bet for any message I've ever sent on a dating platform, or any conversation I've struck up with someone IRL, including people who later became my partners or my wife. I predict this is the case for... almost everyone?
My prediction for Gaza is that no government will reign there. Israel will conduct frequent raids in an on-going basis but based from outside the Strip. Hamas will more or less retain military control of Gaza but by having to operate from underground will not be able to effectively govern the territory. No other force will be willing to operate a government in Gaza. — 75% chance.
The UN and/or Red Crescent will become the de facto government of Gaza. — 20% chance.
Basically there will be no police type activity in Gaza from now on. Just military and basic humanitarian activity.
I'm the #6 most profit on the POTY market, so I think I'm qualified to speak on how I got it right.
The Eras Tour was the highest grossing tour in U.S. history. It generated small booms in excess consumer spending in every city where she performed; According to current estimates, about $4.6 billion in excess consumer spending. It's like Swift ran her own private stimulus program for post-COVID downtowns. If you worked in or near a city she performed in it was all people were talking about for weeks. This is pretty much the largest cultural phenomenon in the US in a very long time.
Meanwhile, most normal people do not know who Sam Altman is, and a minority of them have even heard of ChatGPT.
I think this is a high-information comment and can confirm that Swift is now meaningfully more famous than she has ever been previously.
One issue manifold has always had is with conspiracy theories - people resolving their own markets gets you a bunch of markets like "did the CIA kill JFK" where you're basically betting on how credible the market creator thinks they are, and there's always been an unfortunate amount of those on the site. But with the Gaza stuff in the news it's exploded, with a bunch of highly-active users pushing every sort of antisemitic conspiracy theory under the sun (stuff like "well obviously AIPAC/the IDF/the CIA planted/made up all the evidence for X so this market should resolve no").
Makes it both hard to trust (you now have to look up even market creators of factual-seeming markets to make sure they're not going to misresolve due to conspiracy theories) and makes it a pretty unpleasant experience to go on (especially if you read the comments, you have to wade through a lot of "the *real* terrorists aren't the rape/torture/murder advocates they're the people fighting them).
https://www.axios.com/2023/12/07/palestinian-refugees-egypt-israel-america-relations
>And Israel's previous commitment not to do settlements in Gaza (if maintained) would make a West-Bank-style Gaza better off than the real West Bank.
Why? Did Germany becoming (largely) Judenrein make it better off?
The graph says that Holden Karnofsky is at 50% extinction risk, which I think is based on an incorrect interpretation of the source. The linked source is this post https://www.cold-takes.com/some-additional-detail-on-what-i-mean-by-most-important-century/ which gives 50/50 odds for "civilization could either end entirely, or change so dramatically that 'humans as we know them today' would either not exist anymore, or would at least be a very small part of the population [and] This century is our chance to shape just how this happens".
That's entirely consistent with presently alive biological humans surviving. Biological humans could become a small part of the total population just via the creation of tons of digital minds. (Which humanity may freely choose to do.) So 50% is an upper-bound on extinction risk.
Edit: The 50% is also conditional on PASTA being developed.
Judging by the comment, her performance is a tribute to her stamina and athleticism.