About 21 (ALLFED/Morgan Rivers). Morgan is not actively working at ALLFED anymore (which might be the reason you did not get a response), but the project I think this grant was referring to is mostly finished and has resulted in this peer reviewed paper: https://www.sciencedirect.com/science/article/abs/pii/S2211912424000695
I'm the PI on number 20, microbial biodegredation of plastics, If anyone want to ask questions or comment, then reply to this comment, and I'll try to check my Substack throughout the next few days.
Hey James! I actually just emailed Scott about this and he pointed me at this post. I have a degree in biotech and I'm super interested. Email me at dydavidyoussef@gmail.com
While this is not really my specific field, I attended a presentation by this guy from my university -> https://www.researchgate.net/profile/Alessio-Castagnoli who is currently breeding plastic-producing bacteria (your evil twin, I suppose). What produces plastic (PHA) can presumably degrade it as well. Perhaps he might have useful information for you?
> they expressed concern it was net negative (!) by taking away oxygen and spotlight from potentially more effective orgs
This style of "good thing is bad actually" argument is extremely common, and my general rule of thumb is to dismiss it out of hand unless it is backed up by a very detailed and specific case.
This manner of argument is not necessarily wrong, just nearly impossible to evaluate. It's a favorite argument of ideologues, who will forever criticize real-world progress as half-measures or corrupt compromises, which their preferred solution of course never would be because it would cut through any political economy considerations with a flaming sword of light.
I agree. There's also the related phenomenon where something that exists in the real world right now, warts and all, is compared to an imagined idea, without any warts.
(I often use the phrase "valleys have two sides" when noting instances of this fallacy in conversation, but it doesn't seem to resonate with other people as much as it did with me.)
It's also somewhat silly because it seems to assume that you can a priori figure out which measures are the best. You definitely cannot! Reality is too complex. So the best way to proceed is to make bets, share results, and see what gets adopted and/or funded. Just do things. The world will sort through them. You don't necessarily have to figure out how valuable it all is, apart from deciding how to spend your own time.
Edit: And everything else being equal, more talent and ideas in a field drive more progress, more visibility, and more overall funding. It's not a zero sum game.
It might be interesting to mass-outreach non-grantees asking if there was meaningful progress towards their proposed project even without the ACX grant. In trying to predict successful grants, a false positive rate is very useful, but so would a false negative rate as well. You might be able to adjust towards whatever common factors the successful non-grantees and successful grantees have, while biasing against what the failures had in common.
But it seems like you’ve pulled the same lesson that most VCs have. The founding team is everything, and evaluating the idea itself a-priori only goes so far. Many would-be startup founders who aren’t in any Ivy League or FAANG are bitter about how VCs filter for these things, but it’s simply a matter of picking the best predictor for project success. If you couldn’t succeed in the very straight path towards an Ivy League, how can VCs expect you to succeed in the far murkier path of building a startup?
This gives me the idea of a email response time index, where a person’s manic obsession with doing things NOW is rated by how fast they respond to emails of different types of emails. Maybe a chrome extension with a public ranking?
Agree with the second paragraph ("The founding team is everything"), but I have to add than just because a team is great at one sector (i.e. industrial products) doesn't mean that the same team will be great at a different one (i.e. social media, government). Just look at Musk.
Compared to the average Startup founder, Musk has definitely produced like 10,000x the average in a few different industries. A previously successful exit is the #1 indicator for a good VC bet. On the whole Musk has been in charge of more than one seriously successful company, with a good number of failures.
It's never about causation though, always correlation, as what causes success in business is extremely varied and industry specific and often due to luck. His ability to close a $10+ Billion funding round in a single day makes him an excellent bet for essentially anyone though, as the downside risk, of any of his companies imploding due to running out of money is very unlikely.
We should be very careful to assume a market which underperforms index funds is efficient. Also if it were simply filtering on school quality rather than pure in group preference. we wouldn't expect to see such clustering around specific schools rather than school tier. But we do.
It might be that there's just lower margins in VC to be had in general. While it underperforms index funds, it doesn't do so by a lot, and the variation in returns are significantly higher.
I'm not so sure about school tier. Different schools are known for different things. Yale is known for its humanities and pre-law track, while MIT is known for engineering. They will naturally attract people with different interests that are more or less likely to succeed at a startup.
If you admit allocation by school is not a tier system but by school specialty then you're saying that it's not about whether they could succeed at the path but the result of education. Which is a fundamentally different claim from "getting into an elite school shows inherent traits necessary for success."
And that's without getting into how random admission chances are these days. How much difference is there really between someone who can get into Harvard vs Yale? How many people choose not based on preference but which lets them in?
And VC shows a stronger power law effect than indexes. Effectively, all index funds perform the same as their index. But in VC the majority of firms perform far below average while a few perform far above. If you accept this is true (and it is) that means inherently a significant part of the capital is being inefficiently allocated. That is, all the capital going to funds that turn out to not be top 10% (or whatever the cut off is). So drawing conclusions from where 90% of firms choose to put their capital is a bad heuristic.
Of course, part of the reason that happens is discovery of which VC models work, which appears to be heterogeneous, implying there is not a uniform standard but multiple niches. This is the opposite of what you'd expect in a market with straightforward returns to easily trackable prestige. And in fact implies if you're starting a fund you'd want to differentiate by finding signals other firms have missed. Basically finding your own niche.
There may be multiple causes going into the correlation.
Getting into a top-tier school requires the ability to do hard work diligently, and be quite intelligent.
Among them, perhaps different schools select for different personalities and skills. There's no reason to expect the best writer of the high school class of 2025 should be a good startup founder, and they would probably have a good chance of going to Yale, while there's good reason to expect the best software engineer of the same cohort to get into MIT, and they would have significant advantages in building a startup.
Essentially, two layers of selection effects.
> But in VC the majority of firms perform far below average while a few perform far above. If you accept this is true (and it is) that means inherently a significant part of the capital is being inefficiently allocated.
Not necessarily so. It may simply be that the average returns in VC are lower than the market average, or that investors are willing to accept lower average returns for the possibility of much higher returns on the tails. It may be that investors simply enjoy investing in VC more than an index for reasons of prestige and novelty, so are willing to sacrifice some average returns in return for that.
90% of firms don't use schooling to determine where to allocate capital. The top performing firms use that as an indicator for most first-time founders, among many others. I'm not saying there aren't better indicators out there, but the firms who use this one obviously value it, and they have a much larger financial incentive to understand it than you or I. There may not be better indicators than what they're already using, and those top 10% performing VC firms get there due to luck, or due to their ability to better push their portfolio into subsequent rounds.
How would you distinguish this from simple in group preference? The story that VCs tend to favor the schools they went to or feel most comfortable evaluating people similar to themselves?
If VC returns are below market average structurally then the entire industry is inefficient by definition. There's already plenty of ways to risk adjust returns too. And the claim "90% of firms don't use schooling to determine where to allocate capital" seems insupportable to me. You're really going to claim that heavily recruiting from Harvard is uncommon? Or a secret?
I tend to take VCs at their word about what they're doing which is basically they have a thesis they follow. If that thesis consistently includes "people like me should get large amounts of capital to use with ideally minimal consequences for failure" I do not necessarily take that at face value. Especially if doing that tends to underperform.
Another question: a16z and YC both offer a supportive ecosystem and claim that admission to that ecosystem is a large part of why their founders are successful, rather than innate traits, which means who they select is only part of the cause of success. If this is true, and if schools affect who is best suited to certain kinds of work, do you see how that creates a system where two equally talented people might end up in unequal positions? If it's not true then why do elite VC firms and schools (who presumably are in a better position to know) claim otherwise?
I suspect the email response thing is especially susceptible to Goodharting. In the observational case, people respond quickly to emails because they care deeply and are working obsessively on the problem. In the Goodharted example, people respond because they know their response tendencies are being evaluated. In academia many professors rule their labs like a petty dictator and expect prompt responses to emails at all hours of the day; since the rapid responses are coming from a place of fear, it's not at all clear that the terrified students are more effective scientists (probably the opposite, in fact).
FWIW if I had a few of these "good heuristics for grant effectiveness" I might not even publicize them to preserve their usefulness.
I don't understand the obsession with using email like a dysfunctional instant messaging system. It seems to be a particular generational thing to treat email as a synchronous means of communication. Older people regard it more like snail mail, and younger people wonder why you are using such a lossy and unreliable means of communication for synchronous messaging. If I am collaborating with someone then Discord, Slack, Teams, Zulip, Signal, Telegram, WhatsApp all are more immediate, less mangled through poorly thought out policies, and also less likely to get randomly eaten by an email block list or corporate filter. I think the "good email response latency" signal is strong but only for people roughly aged 25 to 45. Applying it to someone under 25 seems silly, and I would expect any serious person to ask for preferred messaging or to specify a preferred platform unless they only want to talk to Millennials. Is the value of plausible deniability ("I didn't actually ghost you, my spam filter must have eaten your message, but now that a16z invested in you, I miraculously retrieved it from my spam folder") really so high?
Like it or not, email is the only means of communication you can know you share with other people. Not everyone has discord, slack, teams, tulip, signal, telegram or WhatsApp, so if you want to communicate with a stranger “Let’s use this communication method we all already know the other person has” instead of “Let’s us a means of communication that we aren’t sure the other person uses, and may not even be allowed to use because of corporate rules.”
Email also automatically sorts into threads of conversation and is easily group able into temporary group chats through CC and BCC, which you don’t easily get with any of those other methods.
The real question something like that would try to answer is “Does this person have real time notifications, and do they respond immediately if possible?” which loosely represents the singular focus and “Do it now attitude” that is an important component in startup success. If you’re trying to build a company, and you don’t respond to emails after 5 PM, during lunch, or on weekends, you probably shouldn’t be running a startup.
On the topic of climate change activism and plays, I've also come across this seemingly odd intersection.
Several years ago (maybe 2019 or so) I was working for a UK eco-home builder that was building a development of zero-carbon homes in the North of England. Someone asked us for an interview because they were writing a play about a brother and sister who were trying to build a zero-carbon home, where the brother was extremely motivated to tackle climate change. At the time I thought it was strange, but clearly there's a pattern.
I've also been to climate protests where Extinction Rebellion's 'Red Rebel Brigade' put on a kind of street theatre. They call themselves an 'international performance activist troupe'. They're the people who dress up in all red costumes that serve as a calling card for Extinction Rebellion marches. You can see how they'd be a natural fit for theatricality and activism. They are both performing to an audience.
Not that surprising given how much art majors tend to skew left. Sort of a "when all you have is an interpretive dance degree, everything, including carbon-emission induced climate change, looks like a problem that can be solved via interpretive dance" situation.
The Nazis produced a ton of art glorifying their romanticized history and fucked-up worldview. They also had a really strong visual aesthetic. Hugo Boss uniforms, *Triumph of the Will*…there’s no shortage of Nazi cultural icons. It sucks. They ruined an entire branch of aesthetics.
Hey, Trevor here, founder of Highway Pharmaceuticals. First of all, still very grateful for the grant. The money helped, but the credibility also helped a lot.
Second of all, math isn't right on fundraising. Our/ValueBase's valuation isn't public, but ValueBase has publicly *raised* $14 million. It's probably valued at 3-5x that.
Right – I'm not claiming to have come up with iterated selection, but rather iterated selection via direct induction of meiosis. (Although I don't think that idea was novel enough to patent, and it's possible that others had the idea independently.)
It would be cool if you can get it to work with cows, that would be pretty useful (and have lower regulatory barriers than humans).
Current practice is embryo transfer, in which a cow with desirable traits produces the embryo, which is then transferred to donor. I am told that using this method a cow can produce 30 calves per year.
I think they might mean that because they can breed cows for some positive trait (like milk production) faster, cows' milk production could go up an order of magnitude.
Some quick googling suggests that cows did 10x milk production since the middle ages, but largely due to producing milk year round. Further increase would need to increase udder size and/or number of milkings per day, say 2x udder size and 5x the number of milkings - I guess 1 order of magnitude may be possible but two seems unlikely before it becomes either impractical or unethical.
From what Metacelsus and Banko Killdeer comment, I suspect that what they are referring to by productivity here is the productivity of the breeding process. No doubt the end goal is productivity of actual food production, though.
Hmm this seems like a terrible idea in an era of federal dairy subsidies, which are already awful "unaligned agriculture" where they're gaming the fiscal incentives for large producers vs. animal welfare. Unless they'd rather 10x torture (or more) 1/10 the cow population? I suppose it might mitigate methane emissions... although as long as the cheese caves aren't overflowing I suspect they'd keep taking the $$$ regardless. :-(
> Are lobbying organizations a better bet than other types of nonprofit (within the constraints of ACX Grants)?
Yes, development and lobbying are shockingly easy if you just take the time to understand the systems. The system is set up to be responsive!
> One disappointing result was that grants to legibly-credentialled people operating in high-status ways usually did better than betting on small scrappy startups
You admitted you partly selected on this so this conclusion is tainted by selection bias. If you assume two groups have roughly equal percentage of successful people but one group is larger (non-credentialed people) and you likely have worse skill for non-credentialed people by default then selecting a larger share of the smaller group will lead to more successful examples per capita. This is because you'll have more non-credentialed false positives and false negatives in the non-credentialed group as well as fewer false positives in the credentialed group as well as more true positives in the disproportionate credentialed group.
Basically you've rediscovered the social purpose of signaling and concluded that it's better rather than the self-reinforcing mechanism social signaling is. Put another way, you probably included EA credentials as legible because they're legible to you. They are probably less legible to Harvard. This type of thing shows the use is in signaling TO YOU, not the difficulty or generality of the credential, because one of these is literally global and one of them is very niche.
Likewise, any number of niche credentialing organizations create the same effect within their communities because the credential is a signal/coordinating mechanism, not a sign of merit. If it were simply based on merit you'd expect these more niche signaling mechanisms not to work because they'd be strictly inferior to the more global and difficult to achieve ones like Harvard. But instead you see a diversity of credentials with local prestige.
> Someone (I think it might be Paul Graham) once said that they were always surprised how quickly destined-to-be-successful startup founders responded to emails
I think that was Sam Altman (at the time still working in YC):
"ALTMAN: You know, years ago I wrote a little program to look at this, like how quickly our best founders — the founders that run billion-plus companies — answer my emails versus our bad founders. I don’t remember the exact data, but it was mind-blowingly different. It was a difference of minutes versus days on average response times."
Re: climate change, I think spiking demand for energy to power data centers over the next few years is going to require some changes in messaging/strategy to avoid seeming to be standing in the way of progress.
It's mostly cosplay. The game theory of AI means the Sam Altmans of the world will race ahead doing whatever they want while paying lip service to a movement that spends most of it's time spinning up yarns on the internet.
They'll justify maximalist expansion through some combination of fears of China and telling everyone that in the future they won't have to work.
And when that future comes, instead of funding UBI with higher taxes they'll do everything in their power to keep the carried interest loophole and avoid the unspeakable tragedy of paying Clinton era marginal tax rates.
I don't think climate activism is mostly cosplay. US per capita carbon emissions have gone down by almost half even as per capita GDP has increased (see https://www.statista.com/statistics/1049662/fossil-us-carbon-dioxide-emissions-per-person/). This took immense effort, upset a lot of people, and should be considered one of the most surprising and impressive political victories of the past few decades.
I don't think AI safety has been cosplay either. It sounds like your only argument is that maybe AI companies can ignore us; even if that's true, it doesn't make our work "cosplay". But also, most AI companies have safety teams, there's some decent safety research being done, several national governments have AI safety institutes, and there are various AI safety related laws at different stages of completeness. While I agree that our chances are far from certain, I think "cosplay" requires some kind of unseriousness that I deny.
I think the fact that your main worry is whether there will be Clinton era marginal tax rates suggests you're not coming at this from a place of giving a fair shake to our concerns.
I seriously doubt the median climate activist is interested in the technological advances that have lowered carbon emissions, or the fact that we've outsourced a lot of our carbon emissions (and other pollution) to China. The Sunrise Movement spent most of 2021-2022 tying themselves in knots over antiracism nonsense instead of helping Biden pass climate legislation.
An AI safety team to me looks more like a Wall Street compliance department than anything else -- something they do ostensibly to avoid breaking the law but is more geared toward shielding management from liability while making it hard to even *tell* if their breaking the law. Witness how cagey Open AI's CTO was about whether they used YouTube for video training.
The reason I mention Clinton era tax rates is that since 1994 the defining issue for the one percent -- the one thing that they will go to the matt for in Congress, which we see now with the BBB -- is avoid a higher marginal tax rate.
Given the track record, my default position is that talk of UBI is just PR.
Edit: On national governments having AI safety institutes -- the JD Vance argument of "screw that" seems to be winning in the US. Every day I see memes about how the EU is a regulatory disaster.
See this discussion of the claim that we "outsourced our carbon emissions to China" - https://www.noahpinion.blog/p/no-the-us-didnt-outsource-our-carbon . I don't know how you can deny the many carbon use regulations that have impacted every part of life in the US.
I also think you don't understand the role of AI safety teams. They're certainly not "compliance" - so far there are very few regulations to comply with, and many AI companies get along fine without any safety team. If you look at these teams' work, much of it is speculative research about superintelligence - see everything from Anthropic's interpretability research to OpenAI's chain of thought faithfulness and scaleable alignment work. Some of it also has near-term implications but I think the long-term work stands on its own.
Not that it matters, but I think your claim that the "defining issue for the one percent" is tax rates doesn't check out either. The one percent are currently about 50-50 Democrat vs. Republican (see https://catalist.us/whathappened2024/, but this is in the context of a good year for Republicans, and in 2020 it was more Dem-skewed), even though Republicans have consistently been the party of tax-cuts-for-the-rich. I think the top one percent don't really have a single defining issue, but insofar as they do it's mostly culture wars stuff like everyone else.
For all the time and money spent AI safety -- did *anyone* predict that the first casualty would be college students cheating on exams? Have any of the AI safety people suggested legislation or other systemic mechanisms to prevent that? ("Just use bluebooks" is not systemic.) It's like the phones problem. Everyone knows phones are driving us crazy but it's only in the last couple years that individual school districts are doing anything about it. There are more systemic solutions that could be done but everyone knows the lobbying apparatus would prevent it from even getting out of a committee.
As someone who was married to a Chinese woman and spent significant time there, I simply disagree with Noah. It is plainly obvious to anyone who visits China that there is an enormous amount of unreported emissions happening there. Noah has expressed doubts about China's economic reporting in general, so I don't know why he believes them in this one case.
In 2008 the the US Embassy in Beijing started tweeting out daily air pollution readings from their own monitors. They dramatically differed from Beijing's reported numbers (the embassy stopped doing that in March btw).
On the other hand, there is no similar mechanism for carbon emissions. Given how we've seen China (not) handle something that directly leads to emphysema and asthma for their own population, why would we expect them to be vigilant about something that doesn't do have such direct effects?
The 50/50 one percent thing is better understood in terms of the source of their wealth. One faction is the Abigail Disney/Christy Walton class -- heirs who donate to do-gooder orgs like the Sunrise Movement and the Hewlett Foundation. It's like the Ford Foundation -- everything that starts conservative ends up liberal. And they've *hurt* Democrats in terms of electability and being able to wield power. That's what "the groups" discourse is about.
Then there's the Peter Thiel/Ken Griffin faction. They are working every day in their business and can see direct, material results from their efforts to influence tax law and regulation.
Open AI, Google, and Meta all spend a *lot* on lobbying, and their AI safety teams can be viewed as an extension of that. "See? We really care about safety so how about we table legislation for now."
Even if I buy the AI safety stuff you're selling, I still think the game theory means most of it will be for naught. When push comes to shove and governments show they're willing to spend significant sums on mass surveillance, hacking of nation states, and autonomous killer drones, those companies will provide a solution. Theres'a reason Palantir's stock has gone parabolic.
>US per capita carbon emissions have gone down by almost half even as per capita GDP has increased
Nit: At least in the electric power sector, this is mostly from a shift to natural gas, and should probably be considered an outcome of successful fracking rather than of politics.
>Lower CO2 emissions have largely been a result of a shift from coal to natural gas in the electricity generation mix. In 2005, coal made up 50% of U.S. electricity generation; that share declined to 23% in 2019. Conversely, natural gas increased from 19% of total generation in 2005 to 38% in 2019.
I was in Alberta when an interesting ranked choice election happened. It was for leader of the Progressive Conservatives, which was effectively for the Premier of the province. The top two contenders went attack dog on each other, prompting all their voters to put the third place guy as their second choice. So when neither won a majority the third guy cleaned up on the second round
This dynamic could kill excessive partisanship. In this particular instance the third place guy, Ed Stelmach, was not a quality candidate compared to the other two imo. One of the most confidence shaking moments of my life was when he raised the min wage as oil prices crashed and help wanted signs vanished across the province. The guy had no competition and was the leader of a conservative party in a conservative province. He also didn't just neglect to comment on a 'human rights' commission ruling that banned a pastor from public comment about the gay agenda for life, including emails - he vocally supported it
Presumably though, quality candidates would soon learn that excessive partisanship can hand the election to the guy who got 20% in the first round. This does make me thing approval voting is far better even if ranked is a solid upgrade
What you're describing only happens in Condorcet systems. IRV (Instant Runoff Voting) wouldn't lead to that, as the "third place guy" would get eliminated in the first round
The description above is _exactly_ why I prefer IRV. It forces candidates to run on their own merits rather than just attack each other. You actually need significant first-place votes to win
I think Condorcet would sufficiently incentivize positive messaging, since the cost of attack ads is borne by one candidate but the benefit is shared with all the other non-attacked candidates.
The Alberta election seems to have been the result of the campaigns not understanding the incentives, which I would expect to sort itself out in a cycle or two.
Stelmach may objectively suck, but so long as majorities prefer his nonsense to either of the other candidate's he's still the least-bad choice. And without a motivated base of support, his worst policies will likely be comparatively easy to unwind.
no. cardinal methods (score voting and approval voting) are probably better in real life or at least nearly as good, and vastly simpler, cheaper, more politically viable, and better for escaping duopoly.
I think in an ideal world, one of the more complicated voting methods would be ideal… but approval voting seems to me the perfect combo of likely to have an impact, easy to understand and implement, and above all, the absolute least likely to generate voter’s regret! Because if any of your approved candidates win, you’re “happy” with your vote. It does exactly what it says on the tin. Want to protest vote? You can. Like a major party person and also a third party person? Do it. At least on an individual level, it’s empowering. And I think it does enough to at least crack the door to third parties and less polarizing politics.
By contrast I attempted to explain ranked choice voting to my relatively smart parents who actually were in a spot in Oregon trying it out, and basically ended up saying something to the tune of “just rank like they tell you to and trust that it works magically”. But that’s pretty unsatisfying.
Of course if you think moderation and civility are bad, approval voting is not for you. But I think most people like those things?
approval voting is 95% as accurate as the more complicated voting methods and radically more politically viable. seems to me about 20-100 times more impactful than everything else here, or really anything else the EA community is talking about, from a cost-benefit POV.
> Someone (I think it might be Paul Graham) once said that they were always surprised how quickly destined-to-be-successful startup founders responded to emails - sometimes within a single-digit number of minutes regardless of time of day. I used to think of this as mysterious - some sort of psychological trait? Working with these grants has made me think of it as just a straightforward fact of life: some people operate an order of magnitude faster than others.
I know a few people who respond to emails quickly but not very well, and I also feel like this sort of “respond to emails quickly” mentality lines up well with people who don’t take their work-life balance very seriously either—which is probably a good trait to have if you want to be a startup founder.
I’ve been through periods where I respond very quickly and periods where I need reminders to respond. It is absolutely a window into mindset, energy, motivation, dedication and prioritization.
Someone who responds quickly is on top of things. Someone who doesn’t is mentally overwhelmed, distracted, disorganized or depressed.
For reference, the case for instant runoff voting, from that link:
1. A lot of places already use IRV. Approval voting is new, and dangerous.
2. In two case studies that we characterize as using approval voting, of which one actually does use approval voting, people didn't vote for enough candidates. This pathological behavior comes from the fact that approval voting rewards you for restricting your votes; additional votes dilute the power of each vote.
3. However, white voters voted for more candidates than black voters did, giving whites a larger say in the election than blacks got.
4. 51% of voters (in approval elections) were able to correctly describe how it worked.
5. No comment on how many IRV voters were able to correctly describe how it worked.
6. Dartmouth experimented with approval voting, but abandoned it after several student presidents were elected with support from less than 40% of voters.
approval voting is radically superior to IRV and vastly more viable from a cost-benefit standpoint. it passed in fargo and st louis by 64% and 68% respectively.
> In two case studies that we characterize as using approval voting, of which one actually does use approval voting, people didn't vote for enough candidates.
there is no such thing as "enough" candidates. it's perfectly possible to elect the social utility optimizer with everyone approving only one candidate. approval voting has performed fantastically in multiple simulations which incorporate varying percentages of tactical "psychopath" voters, who absolutely "bullet vote" for only one candidate whenever it's the best strategy. so this is not a real issue, it's just a misunderstanding of basic math. lots of discussion by math phd's and other assorted experts here.
> 6. Dartmouth experimented with approval voting, but abandoned it after several student presidents were elected with support from less than 40% of voters.
this has been debunked to death. how have you not done basic research on it?
Maybe you should try reading my response a little more... intelligently?
I didn't say those were my ideas. I said "the case for instant runoff voting, from that link". Those are the points presented in the link Taymon A. Beal provided. In your model of what my comment says, what is the meaning of points (5) and (7)?
As you note, fairvote's "arguments" are stupid and explicitly self-contradictory. I felt this was obvious enough that there was no need to call it out.
> there is no such thing as "enough" candidates. it's perfectly possible to elect the social utility optimizer with everyone approving only one candidate.
On the other hand, that one is easily fairvote's strongest argument. If everyone approves only one candidate, you're doing nothing other than using plurality voting. Reforming plurality voting into more plurality voting is not a worthwhile use of time.
no, approval voting dominates condorcet, in most realistic circumstances involving strategic voting. approval voting always elects the condorcet winner under realistic models of strategic voter behavior, whereas tactical condorcet voters can paradoxically fail to elect the condorcet winner. this is intro level voting theory.
I think that most of those points only work in pretty contrived scenarios, though, and in realistic elections with realistic preference flows and voters who aren't time-travelling wizards they don't really come up.
Did you see the Part II post (https://thegreymatter.substack.com/p/case-against-rcv2) that I linked to at the bottom, where I show a real-world situation? It shows that there was an election spoiler (Wright), the best head-to-head candidate (Montroll) didn't win, some voters would have been better off had they been strategic (Wright voters voting for Montroll instead), and even the weird case where a candidate (Kiss) does better but then loses the election.
Of course, voters aren't going to know exactly how to be strategic to get the outcome they want, but isn't the fact that it's possible a demerit for the election system? Is the fact that I found all of these in one of the few elections of this type somewhat convincing? If not, what do you think would convince you?
The rule of thumb for selecting your cutoff in approval voting is to approve every candidate you like more than the one you expect to finish second. Even the rule of thumb is very complicated to enact because it relies on potentially unavailable information.
I guess two front runners would probably be common even after approval voting is rolled out for some time, because of psychology and limited attention. So most people could just choose every candidate they prefer over the leading candidate on the opposing side (the one they get the most ads for, say).
But if there is no front runner of the opposing ideology, you'd need a prediction market to be in place before you could figure out how to correctly cast your vote. If you don't fit into the two standard sides, it's especially difficult, since you wouldn't have a way of guessing which of the two frontrunners to even consider or use as the benchmark for your ballot.
There's a lot to like here, the anti-mosquito drones in particular look like they'd have both public-health utility and market potential. The AI safety advocacy probably dwarfs anything else going on here in terms of magnitude, it's just hard to know what actions are actually going to be relevant so the existence of advocacy groups that can pivot as the industry changes is really the only way to handle that. (I don't have any background in grants, and am terrible at writing and funding them, and have no worthwhile opinion on whether you should keep doing this, but I do think you continue in some fashion to try to place people in key policymaking spots ahead of the big AI decision points.)
I'm somewhat baffled that the animal welfare team convinced farmers to selectively abort all the male chickens. I eat a bunch of chickens, and eggs, so this is "not my fight" I suppose. I understand that under some value systems this makes sense, but it definitely approaches a danger zone, and I gather there has been some heated discussion of "negative utilitarianism" lately. Chickens don't have future-oriented mental states, perhaps they don't even have preferences at all aside from avoiding pain, in which case this approach could be correct.
I saw a reddit discussion about this and it was suggested that farmers could raise the male chickens for meat, and just don't because they'd have a "surplus" of meat they couldn't sell which ends up going to developing nations, and this outcome was described as being bad because that weakened the local chicken farming economy in those developing nations. That seemed like a rather questionable justification to me, perhaps that ought to be looked into. If I were an activist against factory farming, from a practical purpose I don't know if I'd want to lose "dude they put all the male chickens into something called a grinder a day after they're born" as an argument by just killing them all in-ovo, they just sane-washed factory farming for an outcome that may amount to the same thing.
I'm confused about your objection; even if chickens have subjective experiences, chicken *embryos* probably don't, so I don't see anything wrong with selectively aborting them?
Agreed on the mosquito drones as a visibly interesting approach to a perennial problem. Though…maybe one with some unfortunate consequences. I get that humans don’t output a specific ultrasonic frequency, and aren’t quite as vulnerable to a small electric charge…
Regarding male chickens: I don’t see how aborting male chickens could be any *worse* than the status quo. Even if we assume the zygote gets exactly the same internal experience as the chick would, it just happens sooner.
What exactly is the “danger zone” you have in mind?
The formulation by which it’s a good act to abort them rather then let them be born and live a day and then be killed is one that can lead to bizarre calculations of utility. In very fact-specific individuated cases it may be fine to argue “this person or creature has so much suffering it would have been better to prevent its existence” but I don’t think I agree with it even then, you have to invent pretty extreme hypos for it to make intuitive sense. So I’m suspicious of applying that kind of formulation to an entire class of anything this broad. The one day of life probably isn’t trivial, chickens presumably don’t have a will or self/concept but they are programmed to do a few things, and in that one day they would bust out of their egg, look around, imprint on a nearby creature, maybe that’s a really good chicken-day. “One day of life before you’re tossed into the grinder” even kinda sounds like one of those extreme hypotheticals, and it still seems dubious that nonexistence would be better than existence there.
I would assume that they would lose money raising the male chicks for meat because they would be competing with meat from chickens bred to optimize meat production and wouldn't be able to sell the rooster meat for enough money to pay for the cost of raising them.
Hey CJW! Just wanted to provide some context that may help understand the situation a little better:
- On selling the meat, you're liking referring to something that happened back in the late 2010s. After Germany and France banned the practice of male chick culling, in-ovo sexing technology was still fairly rudimentary at that stage, and not all producers had access to it. As a result, they reared the males and sometimes exported the meat to other countries. The reason it was cheap was because it was a byproduct that they had to get rid of, not because doing this is particularly economical. The other commenters who have pointed out that males of the laying-hen breeds are uneconomical to raise for meat compared to commercial meat chickens are correct. As in-ovo sexing technology further developed pretty much all producers chose to use it rather than raise males for economic reasons. In the US, it's not possible to slaughter significant numbers of males of the layer breed because they're a different size and shape than the typical meat birds we eat, so slaughter facility infrastructure isn't set up for it. Even if it was, it wouldn't be economical, and many people report not liking the quality of the meat anyways since it tends to be tougher than the meat we're used to eating.
- On whether in-ovo sexing amounts to the same thing as maceration, there have actually been a few studies on the cognitive development of chick embryos over the course of the 21 day incubation period. The evidence could always be a better, and ultimately consciousness isn't something I think we fundamentally understand, but generally in-ovo sexing works before there's meaningful evidence that the embryos are conscious or feel pain.
- Activists against factory farming can and have made the argument that you describe, but Innovate Animal Ag isn't that class. We're activists for better animal welfare. Welfare is low enough during day 1 inside a hatchery that it's a clear win in my mind to not experience it, under many ethical theories, not just negative utilitarianism.
Hi, thanks for the reply, I believe that does explain what I had read because it concerned an EU poultry market.
I also find it amazing that the cognitive development of chick embryos has been studied. I still lean towards thinking "existence at all must be better than non-existence even in this case", though there is likely some point in development before which I probably wouldn't consider it the "same" individual, so that could be relevant. If there were a process that made eggs never fertilize with male chicks in the first place, like if they put the roosters' sperm in a centrifuge or something, that would seem better to me if the price could be made comparable. As I said, I do eat chickens so I don't exactly have a moral high ground here, but I guess I still care a nonzero amount. Anyhow thanks for the explanation.
On Grant #14 on the list, as second author on the paper (as an advisor on the project that contributed, but was not paid by the grant,) I strongly second that it cannot be evaluated on its own. That's the problem with marginal work promoting changes to complex systems; even if we get HCTs for the next pandemic, or get more of them for existing diseases, you'd need to do an infeasibly complex Shapley-value analysis across tons of efforts and projects with tons of unknowns to attribute part of the impact to this grant.
But, critically, difficulty measuring impact doesn't function as much evidence about extent of impact, and I think that the paper helps shift the burden of evidence for future studies and thereby saves significantly more lives in expectation than the cost. (I'm very comfortable with that estimate, albeit with low confidence about magnitude.) And I'm grateful that grants like this can be made by funders who can appreciate that it's hard to measure impact, and the expected impact is still the critical factor by which to judge grants - even though subjective estimation of that counterfactual impact remains unavoidable.
Regarding RadVac; I initially read Wentsworth's post on LessWrong about "Making Vaccine".
I went ahead did the same, and administered it to ~40 people, none of whom died of Covid (although in expectation none would have anyway). I also took their 40-page whitepaper and made it into a 2-page policy brief, and created an easy-to-follow instruction page for administration.
I also got involved with the RadVac team, and helped them get their proposal in front of high-ranknig government officials in several countries; ultimately none were willing to sponsor challenge trials, but we knew going in it would be a high-variance / high-expectation long-shot.
I think there's something to be said here about "pulling sideways". It was definitely a lot of effort, but there were some relatively-close almost-successes. It has significantly improved my belief that it's possible to make an impact in labor-limited areas.
> "This one is confusing to evaluate; the specific proposal failed, it encouraged its opponents to create a distraction proposal to sabotage it, and the distraction proposal unexpectedly passed, meaning that Seattle did get a more interesting voting method after all (although unclear whether it’s good). Is this a success of our grant?"
Fun fact: this is also how sex discrimination was banned in the United States (poison pill added to the Civil Rights Act of 1964 that failed to peel off enough votes to kill it), so I think you're in good shape calling this a success.
In short: Howard Smith, who got that part added, may have intended it as a poison pill; but the National Woman's Party, who encouraged him to add it, definitely intended it sincerely (even if perhaps they thought the reason he'd add it was as a poison pill).
(Although, now that I read this Wikipedia article, there seems to be more reason than I thought to believe that it was in fact always meant sincerely and not as a poison pill.)
Regarding #6, I was that climate activist who wrote a rock opera, but it really wasn't a climate activism theme. More a mytho-poetic, sci-fi, rumination on love with some inclusion of terraforming snuck into to the plotline. I do like the idea of using music and theater to capture the public's attention to pressing topics like climate change, and truly do appreciate the link to my opera recording.
61 certainly sounds like something I'd participate in. I seem to be unusually good at learning across a wide variety of areas, so I might provide some unusual data.
Hello - founder of Innovate Animal Ag here! Responding in particular to this: "One thing I still don't understand is that Innovate Animal Ag seemed to genuinely need more funding despite being legibly great and high status - does this screen off a theoretical objection that they don't provide ACX Grants with as much counterfactual impact?"
First of all, thank you so much for your support, and for your kind words. ACX has been extremely helpful to us in many ways including to but not limited to funding.
To you, we may seem "legibly great," but many people don't share your perspective. We have a very different approach from most animal welfare organizations in that we work very collaboratively with the animal ag industry to find win-wins - technologies that are good for welfare but also have a value proposition to farmers. Many from the traditional animal advocacy world are skeptical of this collaborative stance, and some even find it heretical. This, combined with the fact that most donors are still mostly interested in companion animals rather than farm animals, means that ~90% of donors who traditionally fund animal welfare wouldn't fund us.
In other words, it's great to have donors who are weird in similar ways that we are weird. It's possible that we could have raised this money from other sources, but this would have taken a lot longer, and distracted a lot of focus from program work. It also may have pushed in directions that are more legible to traditional funders, but ultimately less impactful.
Another counterfactual benefit of the ACX grants is that we hired multiple people either from ACX classifieds, or who read about us through ACX grants. Again, the reason for this is because your audience is similarly weird to us, and is particularly likely to be excited about our unique theory of change.
My jaw dropped when reading the review, from my perspective you've been astonishingly successful. A truly enormous amount of good done for the money, sincere congratulations.
As a Seattle voter who wanted approval voting and is also uncertain about whether IRV is a net improvement for this specific circumstance: I consider this a successful ACX grant. One of our biggest weaknesses as a civilization right now is status quo bias and specifically a strong bias against experimentation and trying new things. This resulted in a major municipality trying a new thing, which is Good.
Will this change your choice in Beer? If so, why or why not? If it will, how confident are you that anyone else will also change? (If so, you're going to have to overturn a lot of the craft beer lobby in order to get your labelling implemented.)
Unless you only mean to label endocrine disruptors in shampoo? (Which is a very underlabeled product).
Getting instant runoff voting to pass is absolutely a success. Instant runoff is better anyway (Sorry to the approval voting partisans, but they're the one with the fringe view here.) It's probably a low investment lift to get ranked ballots to pass in other places, particularly rural parts of the west coast.
> (Sorry to the approval voting partisans, but they're the one with the fringe view here.)
That's not really an argument, that's pure argument by popularity.
Personally, I think of IRV as being a lot like Esperanto; it's the first alternative that people tend to encounter, and therefore is championed by those who have realized there's a problem but *haven't* realized that there might be more than one approach to a solution and who therefore have stuck to the first one they encountered. Which isn't saying you're wrong, necessarily (although I think you are) -- just that there's an explanation here for popularity here other than it being the best, just as Esperanto isn't the best interlang (on which subject I must of course link to http://jbr.me.uk/ranto/ ).
(Yeah OK I realize that the international language problem that Esperanto is trying to solve is pretty fake because in reality people just use English, but, the comparison still holds IMO :P )
The full argument is a bit involved, but basically approval voting is vastly more subject to gaming than ranked ballots are. The details are technical and beyond what I can explain in a blog comment and in the past when I tried to explain it to approval voting enthusiasts it proved to be a waste of time because they didn't understand the issues and weren't interested in learning.
this could not be more wrong. approval voting is incredibly resistant to tactical voting. see bayesian regret figures from princeton math phd warren smith and voter satisfaction efficiency metrics from harvard stats phd jameson quinn.
there's even a mathematicaly PROOF that approval voting elects the condorcet winner (a pretty good result) even if 100% of voters are maximally strategic.
i would really not preach to people about not being interested in learning, when you could have learned this with 10 seconds of google searching and 5 minutes of reading.
You fundamentally misunderstand that result. It says that if votes are maximally strategic - a difficult thing to do in practice and a very unrealistic assumption - then it will get the condorcet winner. This is in contrast to what happens if they aren't strategic, which is they won't, because of obvious reasons like voters who disapprove of both of the leading candidates having their votes not counted. Requiring voters be strategic is a bad thing. What always gets the condorcet winner regardless of how strategic or non-strategic voters are is ranked choice ballots using a condorcet algorithm.
first of all, i co-founded the center for election science with warren smith, the princeton math phd who created that site. i've contributed to several pages on the site, going back to 2006—including this one on...optimal tactical score/approval voting:
you are demonstrably mistaken about several core voting theory concepts.
misunderstanding: you seem to think it's desirable to elect the condorcet winner. this is false. it's _mathematically proven_ that an electorate may prefer someone else, possibly even the condorcet _loser_. this is one of the most elementary points of the field of voting theory, and largely why arrow's theorem is noteworthy. it's explained here in layman-friendly terms.
the CORRECT measure of voting method performance is "voter satisfaction efficiency" aka "bayesian regret". score voting and approval voting outperform condorcet methods in a wide range of scenarios.
misunderstanding: What always gets the condorcet winner regardless of how strategic or non-strategic voters are is ranked choice ballots using a condorcet algorithm.
utterly false. this is one of the most common novice misunderstandings in the field.
and since most users of ranked voting methods will just assume it works like borda (literally what most people i've spoken with in san francisco and berkeley when i lived there thought), they'll just naively strategically exaggerate anyway. all of this is common knowledge to those of us who've worked in this field for more than a week.
incidentally, you can see computer simulations here showing that, under plausible models of voter strategy, approval voting is BETTER at electing the condorcet winner than actual condorcet methods.
misunderstanding: "Requiring voters [be strategic is a bad thing."
i'm not saying we REQUIRE them to be strategic to get a good result. i'm saying that it's good to have a fairly benign, even GOOD, "worst case scenario" in the event that many of them ARE strategic. because in the real world, people WILL be strategic.
if voters are honest under approval voting, they can do BETTER than getting the condorcet winner, because in many cases, the social utility maximizer won't be the condorcet winner.
Approval voting is not a loophole to Arrow's Theorem. It has the same limitations as ranked choice ballots. I'm done discussing this with you because you have once again demonstrated that you're a crank and interacting with you is a waste of my time.
all the evidence says approval voting is better than instant runoff voting. like in literally every way. more accurate, simpler, precinct summable, cheaper, etc.
“I’d be interested in hearing from research psychologists about whether this project [clearerthinking.org] is well-known in your field, and how effective it actually is about keeping you guys honest.”
I’m a research psychologist and never heard of it. I love the term “importance hacking”. I think that’s a very real phenomenon, but I think they’re missing a dynamic that’s causing it.
The most prestigious journals are the most generalist (eg. Science, Nature). Because these journals get submissions from every possible subfield, their editors are necessarily less expert in each. They judge which papers to send out to review based on how important the paper seems, but they lack the expertise to know, opening the door for hype to be effective. They also don’t know the relevant experts to ask to review and often ask high-profile researchers in the bigger field who are not actually expert in the relevant subfield, and often don’t know that something isn’t entirely original or as important as how it’s being hyped. Some researchers have figured out that if they hype their study well (often not just the result, but also the importance of the question being tested), they’ll actually have an easier review process AND a higher impact publication. Most researchers I know quietly agree that the most important publications in their area are in somewhat more specialist journals, where the editors are area experts and the reviews are tougher.
A feedback mechanism that could counteract this is that those hyped papers should ultimately get cited less by the experts who work most in that subfield, since they’re not actually that important. But publication in high prestige venues tends to break that mechanism. For one, the papers are highly visible and end up being cited outside the subfield by other non-experts who don’t know better either. For another, there’s usually only a couple papers per year published in Science in a given subfield and everyone in that field knows those papers. Both expert and non-expert reviewers will expect you to cite them.
One solution would be to have more editors who are expert in more areas at these journals. But that costs the journals money (unlike reviewers, editors are paid, and at these journals they’re full time editors, not working scientists) and there’s currently no incentive for the journals to do it.
24, per their own description, is trying to make animal agriculture more productive in ways that align with cost efficiency. I don't know that much about bio stuff so am not sure whether they can, but if true, that further entrenches animal agriculture and diminishes the market share for plant-based products. This seems like it's extremely bad from an animal rights perspective, which it appears that you take, and would probably make that grant net negative if there's a sizable chance they actually do anything.
Animal agriculture is often conducted on land that is impractical for plant-based farming. Is your thought that cattle should be driven extinct, or let roam free and wild? (Animal rights cattle raising is to feed them grass, let them live outside -- with a leanto for shelter when it's horrid outside, and otherwise make sure they get pretty happy lives, doing what they'd do in the wild. Then they have one bad day. This is not very different from hunting, if you aren't hunting white-tailed deer (which are destroying our forests).)
24 was originally working on IVG for humans, and we funded them on that basis. They pivoted to cows years later. I suspect the cow work is just a testing ground before they move back to humans. If the cow work does speed up IVG for humans, I'll accept it as a necessary cost.
That Manifund page should continue to work, but fwiw, if you're interested in donating to ACX Grants, I'd suggest holding off for now, and wait for Scott to put out the "lingering questions & next steps" post!
>This team prefers has asked me not to discuss their progress publicly, but you can probably guess what their lives are like right now, and your guess would be correct.
Dumb question by a non american. Are they working on the current pandemic or fired by Trump/DOGE?
Just to say it out loud: This is an astonishing success rate given the very light process (which was the point), congratulations to the grantees, but also Scott and the reviewers.
Iny view, the entire spending is easily justified by 1-2 of the successes, such as Innovate Animal Ag, and there's potential for dramatically more (such as maternal care and lead batteries in Nigeria).
I weight the AI safety stuff at near 0, but that doesn't matter at all here - even aside from the big real world successes, there's a surprising amount of real research and development done here.
If the mosquito drone guys ever want to expand operations, they should look into making anti-invasive species drones for the New Zealand government. The NZ government has a "plan" to be rat-free by 2050 (I think that's the year) but it will require new tech.
> 4: Alice Evans’ Research On “The Great Gender Divergence”
> Meanwhile, Latin America and the Caribbean face a different obstacle: pervasive violence elevates femicides.
Androcides outnumber femicides almost everywhere in the world; in no country in Latin America or the Caribbean are women more than 35% of all homicide victims (40% if we include Guyana, which is South American but not Latin). https://en.wikipedia.org/wiki/Homicide_statistics_by_gender
It doesn't, but (as I thought was obvious) my point is an increase in femicides (is that what "elevates" means in this context?) isn't gender inequality/divergence, which was the context.
> the distraction proposal unexpectedly passed, meaning that Seattle did get a more interesting voting method after all (although unclear whether it’s good). Is this a “success” of our grant?
I think not for two reasons. The first is that IRV has sufficiently different properties than AV that I think the two voting systems achieve fundamentally different goals. (AV avoids center-squeeze, IRV does not. AV has a lot of nice monotonicity properties, IRV does not, etc.). The second reason is more of a quibble, which is that I think more needs to go into entrenching whatever voting system is established, whether that be legally or by means of raising public support. IRV systems are frequently repealed (sometimes by voters, whereas, in the US, AV has only been repealed at the state level in spite of voter support) even after initiatives pass, and I'm not sure the Seattle case should be expected to last.
Hello, I work on the frontend for Baby Steps (#60, helping prospective parents choose IVF clinics with a success rate prediction model). Thanks so much for the grant! Sorry I didn't thank you in person while I was in Berkeley.
If anyone has any questions about it, I can answer them in this thread :)
Thank you, Scott! Without you or the grant competition, we would never have built Manifold.
Even the idea of user-created prediction markets was something we thought of a few months before ACX grants while reading your Mantic Monday series.
I knew I wanted to work with Austin, having met him online through Discord. And the grant competition was the first thing we decided to collaborate on. We brainstormed ideas and landed on user-created prediction markets (on Solana haha).
Austin flew to meet my brother and I, and the next day was coincidentally our final round interview for YC. We pitched what would become Manifold in a Zoom call on our living room couch. It was basically the first time the three of us ever met. They didn't accept us though, haha. They didn't like our two-week old crypto idea, but that didn't slow us down.
It was December 2nd. We worked every day until Christmas to build the first version of Manifold. After a call with Zvi Mowshowitz, who was very negative on crypto, we came up with the pure play money idea, and built it out.
It was ready the day Scott emailed us to say we won. Scott posted the link in the ACX winners post; we instantly got hundreds of users. The rest is history.
This is Alene from #25 (Legal Impact for Chickens).
So grateful to ACX and all the ACX readers who have supported LIC!
Also since we sent in our update, we have had a new victory: LIC proved in court that SPCAs can sue to stop cruelty to farmed animals. California's Humboldt County Superior Court just ruled that LIC's cruelty lawsuit can proceed against well-known dairy operation Alexandre Family Farm, which supplies Whole Foods.
California has a law against animal cruelty—which extends to farmed animals. But the fact that prosecutors rarely enforce cruelty laws against the powerful corporate interests that control animal agriculture leads some people to falsely believe that animal ag is above the law.
The court's order determined that LIC's cause of action, enforcing the cruelty law against an agricultural operation, is valid. The court ruled that an organization like LIC can directly sue ag companies for cruelty, without needing to pass any additional bar for standing.
We're excited to move forward with the facts stage of the case. And we hope other California-based organizations will start to bring other cases like this for farmed animals, too.
Congrats, and I'm glad about the work you're doing! If you spot any ways to force companies to try harder at preventing coccidiosis and salmonella infections I think that could have a very large impact, because I recently took in some rescue hens with coccidiosis and the improvement in their happiness once they were treated for it was quite large. Also if these diseases are prevented rather than managed it may slow down the development of drug resistance.
These are great points. Thank you so much, Catmint. And thank you for rescuing birds!!! The last USDA administration did put out a call for comments about preventing salmonella from getting into the food people eat. And LIC commented asking for solutions that would improve animal welfare, and prevent birds from getting salmonella in the first place. https://www.legalimpactforchickens.org/salmonella . But sadly, the current administration withdrew the proposed rule. :-( Still, there are probably solutions that don't involve the government at all . . . Like maybe individual farmers could simply choose to vaccinate birds for coccidiosis and salmonella? I don't know anything about whether farmers already do that, or whether they could. But I will ask around.
Vaccines would need to be used carefully, because the only scalable options we have right now (as far as I can tell from my brief research) are live vaccines, and those come with a risk of introducing the disease to a flock or a building that did not already have it. But they're definitely very useful for places that do, which unfortunately is most of them. Probiotics apparently help, but I'm not sure how much. There's also some bacteriophages that attack salmonella, which sounds very promising, but as far as I can tell this is still an active area of research rather than a commercially available product. (This talks about both: https://pmc.ncbi.nlm.nih.gov/articles/PMC10672927/) LIC's suggestions of decreased flock size and earlier testing also sound effective. Glad you guys already looked into it.
Hi Scott! Longtime fan of your writing here. As it matches this ACX Grants article's topics of "climate activism without an associated musical theater component," and " lobbying/advocacy organizations," I'd love to share what I'm working on.
Since 2024, in addition to my personal Substack The Weekly Anthropocene (https://sammatey.substack.com/, thanks again for linking to my interview with Ozy Brennan!), I've been writing Your Dose of Climate Hope (https://climateactapp.substack.com/), a Substack partnered with the new Climate Action Now advocacy software startup that's providing daily actions to advance hopeful, sensible, incrementalist, life-improving, innovation-incentivizing, pro-permitting reform, and pro-Abundance Agenda climate policy. In many cases, I'm pretty sure that Your Dose of Climate Hope is the only advocacy entity trying to contact federal, state, or city governments on a particular issue, especially our regular feature of highlighting a promising emerging technology (e.g. advanced geothermal, modular housing, AI-enhanced crop breeding, one-dose malaria vaccines, ocean alkalinity enhancement, and much more) and asking policymakers to ensure a supportive legal, regulatory, and permitting process. I think in many cases our letters are going to be the first that a given politician has ever heard of the topic, and we're trying to get in there early with a first impression to show that constituents want to build this innovation.
If you find this interesting, I'm not looking for money, but a Substack Recommendation from Astral Codex Ten would be INCREDIBLY valuable - the Recommendation function has been the primary way I've grown my audience for both projects!
I would posit that a Substack Recommendation amounts to ~2 minutes of your time to directly help advance the epistemic state of climate activism and indirectly help positively shift U.S. state and federal policies.
Hi! This is Michael Todhunter, the guy from #32. Promega indeed offers robotic lab hardware - as does Hamilton, Tecan, Beckman Coulter, and a bunch of other companies. What makes this grant proposal different is that I'm trying to make, for lack of a better term, a self-driving biologist - a robotic system that can plan, execute, and interpret its own experiments. We might buy stuff from Promega, but we're not competing with Promega. FutureHouse is working on similar projects, but I'm focusing on embodying an automated scientist in a robotic lab, whereas, to the best of my knowledge, they're focusing on LLM-based scientific reasoning.
> I'm most ambivalent about our science grants. Many of them say they are successful and can point to published papers which explain the science they did. But it's hard to judge whether anything useful has changed based on the science getting done. I know it's important to fund basic research and not just last-mile technology startups, but it's hard for a mini-grants program like this one to evaluate these kinds of abstract interventions.
I can think of a couple of reasons for this.
1) Science is just really hard *and* really expensive. A single (reasonable) publication can costs hundreds of thousands of dollars to produce. Even just one grad student can cost $75-100k per year when all costs are factored in. Plus we already have systems in place to review and fund science. They aren't perfect, but it usually means there's unlikely to be any low-hanging fruit to be picked unless perhaps if you have a very niche interest.
2) Grant proposals are usually reviewed by experts in the field with many years (often decades) of experience. If a person submits a proposal to you, it likely means they are unable (or expect to be unable) to get it funded through the normal channels and are hoping you and your team will be less skeptical than experienced, field-specific reviewers. This doesn't mean the idea is bad per se, but at least it may not be fully formed yet.
What some foundations do is fund projects at an early stage in order to generate data for proposals to federal agencies. They will make a second year's funding contingent on submission of an NIH proposal, for example. This seems like a better way to leverage a small amount of money to have a bigger impact.
About 21 (ALLFED/Morgan Rivers). Morgan is not actively working at ALLFED anymore (which might be the reason you did not get a response), but the project I think this grant was referring to is mostly finished and has resulted in this peer reviewed paper: https://www.sciencedirect.com/science/article/abs/pii/S2211912424000695
Thank you!
Iirc the first round of grants was completely assigned by you, and for the second round you did a partnership with Manifund to do impact markets.
How is that experiment going?
That's a good question. I'm only evaluating grants assigned by me. I'll ask Manifund how they're handling records for their impact market grants.
I'm the PI on number 20, microbial biodegredation of plastics, If anyone want to ask questions or comment, then reply to this comment, and I'll try to check my Substack throughout the next few days.
Hey James! I actually just emailed Scott about this and he pointed me at this post. I have a degree in biotech and I'm super interested. Email me at dydavidyoussef@gmail.com
While this is not really my specific field, I attended a presentation by this guy from my university -> https://www.researchgate.net/profile/Alessio-Castagnoli who is currently breeding plastic-producing bacteria (your evil twin, I suppose). What produces plastic (PHA) can presumably degrade it as well. Perhaps he might have useful information for you?
> they expressed concern it was net negative (!) by taking away oxygen and spotlight from potentially more effective orgs
This style of "good thing is bad actually" argument is extremely common, and my general rule of thumb is to dismiss it out of hand unless it is backed up by a very detailed and specific case.
This manner of argument is not necessarily wrong, just nearly impossible to evaluate. It's a favorite argument of ideologues, who will forever criticize real-world progress as half-measures or corrupt compromises, which their preferred solution of course never would be because it would cut through any political economy considerations with a flaming sword of light.
I agree. There's also the related phenomenon where something that exists in the real world right now, warts and all, is compared to an imagined idea, without any warts.
https://en.wikipedia.org/wiki/Nirvana_fallacy
https://slatestarcodex.com/2013/03/13/valleys-have-two-sides/
(I often use the phrase "valleys have two sides" when noting instances of this fallacy in conversation, but it doesn't seem to resonate with other people as much as it did with me.)
It's also somewhat silly because it seems to assume that you can a priori figure out which measures are the best. You definitely cannot! Reality is too complex. So the best way to proceed is to make bets, share results, and see what gets adopted and/or funded. Just do things. The world will sort through them. You don't necessarily have to figure out how valuable it all is, apart from deciding how to spend your own time.
Edit: And everything else being equal, more talent and ideas in a field drive more progress, more visibility, and more overall funding. It's not a zero sum game.
It might be interesting to mass-outreach non-grantees asking if there was meaningful progress towards their proposed project even without the ACX grant. In trying to predict successful grants, a false positive rate is very useful, but so would a false negative rate as well. You might be able to adjust towards whatever common factors the successful non-grantees and successful grantees have, while biasing against what the failures had in common.
But it seems like you’ve pulled the same lesson that most VCs have. The founding team is everything, and evaluating the idea itself a-priori only goes so far. Many would-be startup founders who aren’t in any Ivy League or FAANG are bitter about how VCs filter for these things, but it’s simply a matter of picking the best predictor for project success. If you couldn’t succeed in the very straight path towards an Ivy League, how can VCs expect you to succeed in the far murkier path of building a startup?
This gives me the idea of a email response time index, where a person’s manic obsession with doing things NOW is rated by how fast they respond to emails of different types of emails. Maybe a chrome extension with a public ranking?
Agree with the second paragraph ("The founding team is everything"), but I have to add than just because a team is great at one sector (i.e. industrial products) doesn't mean that the same team will be great at a different one (i.e. social media, government). Just look at Musk.
Compared to the average Startup founder, Musk has definitely produced like 10,000x the average in a few different industries. A previously successful exit is the #1 indicator for a good VC bet. On the whole Musk has been in charge of more than one seriously successful company, with a good number of failures.
It's never about causation though, always correlation, as what causes success in business is extremely varied and industry specific and often due to luck. His ability to close a $10+ Billion funding round in a single day makes him an excellent bet for essentially anyone though, as the downside risk, of any of his companies imploding due to running out of money is very unlikely.
We should be very careful to assume a market which underperforms index funds is efficient. Also if it were simply filtering on school quality rather than pure in group preference. we wouldn't expect to see such clustering around specific schools rather than school tier. But we do.
It might be that there's just lower margins in VC to be had in general. While it underperforms index funds, it doesn't do so by a lot, and the variation in returns are significantly higher.
I'm not so sure about school tier. Different schools are known for different things. Yale is known for its humanities and pre-law track, while MIT is known for engineering. They will naturally attract people with different interests that are more or less likely to succeed at a startup.
If you admit allocation by school is not a tier system but by school specialty then you're saying that it's not about whether they could succeed at the path but the result of education. Which is a fundamentally different claim from "getting into an elite school shows inherent traits necessary for success."
And that's without getting into how random admission chances are these days. How much difference is there really between someone who can get into Harvard vs Yale? How many people choose not based on preference but which lets them in?
And VC shows a stronger power law effect than indexes. Effectively, all index funds perform the same as their index. But in VC the majority of firms perform far below average while a few perform far above. If you accept this is true (and it is) that means inherently a significant part of the capital is being inefficiently allocated. That is, all the capital going to funds that turn out to not be top 10% (or whatever the cut off is). So drawing conclusions from where 90% of firms choose to put their capital is a bad heuristic.
Of course, part of the reason that happens is discovery of which VC models work, which appears to be heterogeneous, implying there is not a uniform standard but multiple niches. This is the opposite of what you'd expect in a market with straightforward returns to easily trackable prestige. And in fact implies if you're starting a fund you'd want to differentiate by finding signals other firms have missed. Basically finding your own niche.
There may be multiple causes going into the correlation.
Getting into a top-tier school requires the ability to do hard work diligently, and be quite intelligent.
Among them, perhaps different schools select for different personalities and skills. There's no reason to expect the best writer of the high school class of 2025 should be a good startup founder, and they would probably have a good chance of going to Yale, while there's good reason to expect the best software engineer of the same cohort to get into MIT, and they would have significant advantages in building a startup.
Essentially, two layers of selection effects.
> But in VC the majority of firms perform far below average while a few perform far above. If you accept this is true (and it is) that means inherently a significant part of the capital is being inefficiently allocated.
Not necessarily so. It may simply be that the average returns in VC are lower than the market average, or that investors are willing to accept lower average returns for the possibility of much higher returns on the tails. It may be that investors simply enjoy investing in VC more than an index for reasons of prestige and novelty, so are willing to sacrifice some average returns in return for that.
90% of firms don't use schooling to determine where to allocate capital. The top performing firms use that as an indicator for most first-time founders, among many others. I'm not saying there aren't better indicators out there, but the firms who use this one obviously value it, and they have a much larger financial incentive to understand it than you or I. There may not be better indicators than what they're already using, and those top 10% performing VC firms get there due to luck, or due to their ability to better push their portfolio into subsequent rounds.
How would you distinguish this from simple in group preference? The story that VCs tend to favor the schools they went to or feel most comfortable evaluating people similar to themselves?
If VC returns are below market average structurally then the entire industry is inefficient by definition. There's already plenty of ways to risk adjust returns too. And the claim "90% of firms don't use schooling to determine where to allocate capital" seems insupportable to me. You're really going to claim that heavily recruiting from Harvard is uncommon? Or a secret?
I tend to take VCs at their word about what they're doing which is basically they have a thesis they follow. If that thesis consistently includes "people like me should get large amounts of capital to use with ideally minimal consequences for failure" I do not necessarily take that at face value. Especially if doing that tends to underperform.
Another question: a16z and YC both offer a supportive ecosystem and claim that admission to that ecosystem is a large part of why their founders are successful, rather than innate traits, which means who they select is only part of the cause of success. If this is true, and if schools affect who is best suited to certain kinds of work, do you see how that creates a system where two equally talented people might end up in unequal positions? If it's not true then why do elite VC firms and schools (who presumably are in a better position to know) claim otherwise?
I suspect the email response thing is especially susceptible to Goodharting. In the observational case, people respond quickly to emails because they care deeply and are working obsessively on the problem. In the Goodharted example, people respond because they know their response tendencies are being evaluated. In academia many professors rule their labs like a petty dictator and expect prompt responses to emails at all hours of the day; since the rapid responses are coming from a place of fear, it's not at all clear that the terrified students are more effective scientists (probably the opposite, in fact).
FWIW if I had a few of these "good heuristics for grant effectiveness" I might not even publicize them to preserve their usefulness.
I don't understand the obsession with using email like a dysfunctional instant messaging system. It seems to be a particular generational thing to treat email as a synchronous means of communication. Older people regard it more like snail mail, and younger people wonder why you are using such a lossy and unreliable means of communication for synchronous messaging. If I am collaborating with someone then Discord, Slack, Teams, Zulip, Signal, Telegram, WhatsApp all are more immediate, less mangled through poorly thought out policies, and also less likely to get randomly eaten by an email block list or corporate filter. I think the "good email response latency" signal is strong but only for people roughly aged 25 to 45. Applying it to someone under 25 seems silly, and I would expect any serious person to ask for preferred messaging or to specify a preferred platform unless they only want to talk to Millennials. Is the value of plausible deniability ("I didn't actually ghost you, my spam filter must have eaten your message, but now that a16z invested in you, I miraculously retrieved it from my spam folder") really so high?
Like it or not, email is the only means of communication you can know you share with other people. Not everyone has discord, slack, teams, tulip, signal, telegram or WhatsApp, so if you want to communicate with a stranger “Let’s use this communication method we all already know the other person has” instead of “Let’s us a means of communication that we aren’t sure the other person uses, and may not even be allowed to use because of corporate rules.”
Email also automatically sorts into threads of conversation and is easily group able into temporary group chats through CC and BCC, which you don’t easily get with any of those other methods.
The real question something like that would try to answer is “Does this person have real time notifications, and do they respond immediately if possible?” which loosely represents the singular focus and “Do it now attitude” that is an important component in startup success. If you’re trying to build a company, and you don’t respond to emails after 5 PM, during lunch, or on weekends, you probably shouldn’t be running a startup.
On the topic of climate change activism and plays, I've also come across this seemingly odd intersection.
Several years ago (maybe 2019 or so) I was working for a UK eco-home builder that was building a development of zero-carbon homes in the North of England. Someone asked us for an interview because they were writing a play about a brother and sister who were trying to build a zero-carbon home, where the brother was extremely motivated to tackle climate change. At the time I thought it was strange, but clearly there's a pattern.
I've also been to climate protests where Extinction Rebellion's 'Red Rebel Brigade' put on a kind of street theatre. They call themselves an 'international performance activist troupe'. They're the people who dress up in all red costumes that serve as a calling card for Extinction Rebellion marches. You can see how they'd be a natural fit for theatricality and activism. They are both performing to an audience.
Not that surprising given how much art majors tend to skew left. Sort of a "when all you have is an interpretive dance degree, everything, including carbon-emission induced climate change, looks like a problem that can be solved via interpretive dance" situation.
Conversely, people who feel really strongly about most anything are more likely to produce art about it. (Citation needed.)
Where's all the Nazi art?
They actually have it at a museum pretty close to me! https://kimbellart.org/modern-art-politics-germany
The Nazis produced a ton of art glorifying their romanticized history and fucked-up worldview. They also had a really strong visual aesthetic. Hugo Boss uniforms, *Triumph of the Will*…there’s no shortage of Nazi cultural icons. It sucks. They ruined an entire branch of aesthetics.
Hey, Trevor here, founder of Highway Pharmaceuticals. First of all, still very grateful for the grant. The money helped, but the credibility also helped a lot.
Second of all, math isn't right on fundraising. Our/ValueBase's valuation isn't public, but ValueBase has publicly *raised* $14 million. It's probably valued at 3-5x that.
Good point on ValueBase. I won't ask for your nonpublic information, but regarding your company I thought the "valuation cap" on https://kingscrowd.com/highway-pharmaceuticals-on-wefunder-2022/ was the right number to use, am I misunderstanding?
Oh yeah, that was correct as of that date.
"our technology will increase breeding rates by 10-100x, [making cows orders of magnitude more productive] in a couple of years.”
Dumb question: this sounds like this is expecting cows to produce 10-100 calves per year. That's not possible, so what do they actually mean by this?
It sounds like Jeff wants to do iterated meiotic selection (which I proposed in 2022: https://denovo.substack.com/p/meiosis-is-all-you-need ) to speed up the effective generation time.
Thanks, that makes sense.
I submitted an up
The idea of iterated selection goes back really far:
Georges, M., and J. M. Massey. 1991. Velogenetics, or the
synergistic use of marker assisted selection and germ-line
manipulation
Right – I'm not claiming to have come up with iterated selection, but rather iterated selection via direct induction of meiosis. (Although I don't think that idea was novel enough to patent, and it's possible that others had the idea independently.)
It would be cool if you can get it to work with cows, that would be pretty useful (and have lower regulatory barriers than humans).
Current practice is embryo transfer, in which a cow with desirable traits produces the embryo, which is then transferred to donor. I am told that using this method a cow can produce 30 calves per year.
I think they might mean that because they can breed cows for some positive trait (like milk production) faster, cows' milk production could go up an order of magnitude.
Some quick googling suggests that cows did 10x milk production since the middle ages, but largely due to producing milk year round. Further increase would need to increase udder size and/or number of milkings per day, say 2x udder size and 5x the number of milkings - I guess 1 order of magnitude may be possible but two seems unlikely before it becomes either impractical or unethical.
From what Metacelsus and Banko Killdeer comment, I suspect that what they are referring to by productivity here is the productivity of the breeding process. No doubt the end goal is productivity of actual food production, though.
Hmm this seems like a terrible idea in an era of federal dairy subsidies, which are already awful "unaligned agriculture" where they're gaming the fiscal incentives for large producers vs. animal welfare. Unless they'd rather 10x torture (or more) 1/10 the cow population? I suppose it might mitigate methane emissions... although as long as the cheese caves aren't overflowing I suspect they'd keep taking the $$$ regardless. :-(
> Are lobbying organizations a better bet than other types of nonprofit (within the constraints of ACX Grants)?
Yes, development and lobbying are shockingly easy if you just take the time to understand the systems. The system is set up to be responsive!
> One disappointing result was that grants to legibly-credentialled people operating in high-status ways usually did better than betting on small scrappy startups
You admitted you partly selected on this so this conclusion is tainted by selection bias. If you assume two groups have roughly equal percentage of successful people but one group is larger (non-credentialed people) and you likely have worse skill for non-credentialed people by default then selecting a larger share of the smaller group will lead to more successful examples per capita. This is because you'll have more non-credentialed false positives and false negatives in the non-credentialed group as well as fewer false positives in the credentialed group as well as more true positives in the disproportionate credentialed group.
Basically you've rediscovered the social purpose of signaling and concluded that it's better rather than the self-reinforcing mechanism social signaling is. Put another way, you probably included EA credentials as legible because they're legible to you. They are probably less legible to Harvard. This type of thing shows the use is in signaling TO YOU, not the difficulty or generality of the credential, because one of these is literally global and one of them is very niche.
Likewise, any number of niche credentialing organizations create the same effect within their communities because the credential is a signal/coordinating mechanism, not a sign of merit. If it were simply based on merit you'd expect these more niche signaling mechanisms not to work because they'd be strictly inferior to the more global and difficult to achieve ones like Harvard. But instead you see a diversity of credentials with local prestige.
> Someone (I think it might be Paul Graham) once said that they were always surprised how quickly destined-to-be-successful startup founders responded to emails
I think that was Sam Altman (at the time still working in YC):
https://conversationswithtyler.com/episodes/sam-altman/
"ALTMAN: You know, years ago I wrote a little program to look at this, like how quickly our best founders — the founders that run billion-plus companies — answer my emails versus our bad founders. I don’t remember the exact data, but it was mind-blowingly different. It was a difference of minutes versus days on average response times."
Re: climate change, I think spiking demand for energy to power data centers over the next few years is going to require some changes in messaging/strategy to avoid seeming to be standing in the way of progress.
CMV: AI Safety is the Climate Activism of rationalists.
You'll have to explain more about what you mean.
It's mostly cosplay. The game theory of AI means the Sam Altmans of the world will race ahead doing whatever they want while paying lip service to a movement that spends most of it's time spinning up yarns on the internet.
They'll justify maximalist expansion through some combination of fears of China and telling everyone that in the future they won't have to work.
And when that future comes, instead of funding UBI with higher taxes they'll do everything in their power to keep the carried interest loophole and avoid the unspeakable tragedy of paying Clinton era marginal tax rates.
I don't think climate activism is mostly cosplay. US per capita carbon emissions have gone down by almost half even as per capita GDP has increased (see https://www.statista.com/statistics/1049662/fossil-us-carbon-dioxide-emissions-per-person/). This took immense effort, upset a lot of people, and should be considered one of the most surprising and impressive political victories of the past few decades.
I don't think AI safety has been cosplay either. It sounds like your only argument is that maybe AI companies can ignore us; even if that's true, it doesn't make our work "cosplay". But also, most AI companies have safety teams, there's some decent safety research being done, several national governments have AI safety institutes, and there are various AI safety related laws at different stages of completeness. While I agree that our chances are far from certain, I think "cosplay" requires some kind of unseriousness that I deny.
I think the fact that your main worry is whether there will be Clinton era marginal tax rates suggests you're not coming at this from a place of giving a fair shake to our concerns.
I seriously doubt the median climate activist is interested in the technological advances that have lowered carbon emissions, or the fact that we've outsourced a lot of our carbon emissions (and other pollution) to China. The Sunrise Movement spent most of 2021-2022 tying themselves in knots over antiracism nonsense instead of helping Biden pass climate legislation.
An AI safety team to me looks more like a Wall Street compliance department than anything else -- something they do ostensibly to avoid breaking the law but is more geared toward shielding management from liability while making it hard to even *tell* if their breaking the law. Witness how cagey Open AI's CTO was about whether they used YouTube for video training.
The reason I mention Clinton era tax rates is that since 1994 the defining issue for the one percent -- the one thing that they will go to the matt for in Congress, which we see now with the BBB -- is avoid a higher marginal tax rate.
Given the track record, my default position is that talk of UBI is just PR.
Edit: On national governments having AI safety institutes -- the JD Vance argument of "screw that" seems to be winning in the US. Every day I see memes about how the EU is a regulatory disaster.
See this discussion of the claim that we "outsourced our carbon emissions to China" - https://www.noahpinion.blog/p/no-the-us-didnt-outsource-our-carbon . I don't know how you can deny the many carbon use regulations that have impacted every part of life in the US.
I also think you don't understand the role of AI safety teams. They're certainly not "compliance" - so far there are very few regulations to comply with, and many AI companies get along fine without any safety team. If you look at these teams' work, much of it is speculative research about superintelligence - see everything from Anthropic's interpretability research to OpenAI's chain of thought faithfulness and scaleable alignment work. Some of it also has near-term implications but I think the long-term work stands on its own.
Not that it matters, but I think your claim that the "defining issue for the one percent" is tax rates doesn't check out either. The one percent are currently about 50-50 Democrat vs. Republican (see https://catalist.us/whathappened2024/, but this is in the context of a good year for Republicans, and in 2020 it was more Dem-skewed), even though Republicans have consistently been the party of tax-cuts-for-the-rich. I think the top one percent don't really have a single defining issue, but insofar as they do it's mostly culture wars stuff like everyone else.
For all the time and money spent AI safety -- did *anyone* predict that the first casualty would be college students cheating on exams? Have any of the AI safety people suggested legislation or other systemic mechanisms to prevent that? ("Just use bluebooks" is not systemic.) It's like the phones problem. Everyone knows phones are driving us crazy but it's only in the last couple years that individual school districts are doing anything about it. There are more systemic solutions that could be done but everyone knows the lobbying apparatus would prevent it from even getting out of a committee.
As someone who was married to a Chinese woman and spent significant time there, I simply disagree with Noah. It is plainly obvious to anyone who visits China that there is an enormous amount of unreported emissions happening there. Noah has expressed doubts about China's economic reporting in general, so I don't know why he believes them in this one case.
In 2008 the the US Embassy in Beijing started tweeting out daily air pollution readings from their own monitors. They dramatically differed from Beijing's reported numbers (the embassy stopped doing that in March btw).
https://www.pnas.org/doi/10.1073/pnas.2201092119
On the one hand, it did lead China to do something about the pollution, but that was also in part because there were widespread protests (e.g. https://www.wikiwand.com/en/articles/Shifang_protest).
On the other hand, there is no similar mechanism for carbon emissions. Given how we've seen China (not) handle something that directly leads to emphysema and asthma for their own population, why would we expect them to be vigilant about something that doesn't do have such direct effects?
The 50/50 one percent thing is better understood in terms of the source of their wealth. One faction is the Abigail Disney/Christy Walton class -- heirs who donate to do-gooder orgs like the Sunrise Movement and the Hewlett Foundation. It's like the Ford Foundation -- everything that starts conservative ends up liberal. And they've *hurt* Democrats in terms of electability and being able to wield power. That's what "the groups" discourse is about.
Then there's the Peter Thiel/Ken Griffin faction. They are working every day in their business and can see direct, material results from their efforts to influence tax law and regulation.
Open AI, Google, and Meta all spend a *lot* on lobbying, and their AI safety teams can be viewed as an extension of that. "See? We really care about safety so how about we table legislation for now."
Even if I buy the AI safety stuff you're selling, I still think the game theory means most of it will be for naught. When push comes to shove and governments show they're willing to spend significant sums on mass surveillance, hacking of nation states, and autonomous killer drones, those companies will provide a solution. Theres'a reason Palantir's stock has gone parabolic.
And they definitely will not fund UBI.
Worse yet, the left actively opposed the so-called dirty deal that would have allowed Biden's clean energy program to actually get built.
The AI safety community’s version of that is to oppose all regulation and make fun of Europe for even trying.
>US per capita carbon emissions have gone down by almost half even as per capita GDP has increased
Nit: At least in the electric power sector, this is mostly from a shift to natural gas, and should probably be considered an outcome of successful fracking rather than of politics.
https://www.eia.gov/todayinenergy/detail.php?id=48296
>Lower CO2 emissions have largely been a result of a shift from coal to natural gas in the electricity generation mix. In 2005, coal made up 50% of U.S. electricity generation; that share declined to 23% in 2019. Conversely, natural gas increased from 19% of total generation in 2005 to 38% in 2019.
Liable to remain fringe until one tribe realizes they can hollow it out, wear it as a skinsuit, and build support for an unrelated political battle?
The battle will be against taxation for UBI. Billionaires and VCs need that money to make moar AI.
I was in Alberta when an interesting ranked choice election happened. It was for leader of the Progressive Conservatives, which was effectively for the Premier of the province. The top two contenders went attack dog on each other, prompting all their voters to put the third place guy as their second choice. So when neither won a majority the third guy cleaned up on the second round
This dynamic could kill excessive partisanship. In this particular instance the third place guy, Ed Stelmach, was not a quality candidate compared to the other two imo. One of the most confidence shaking moments of my life was when he raised the min wage as oil prices crashed and help wanted signs vanished across the province. The guy had no competition and was the leader of a conservative party in a conservative province. He also didn't just neglect to comment on a 'human rights' commission ruling that banned a pastor from public comment about the gay agenda for life, including emails - he vocally supported it
Presumably though, quality candidates would soon learn that excessive partisanship can hand the election to the guy who got 20% in the first round. This does make me thing approval voting is far better even if ranked is a solid upgrade
What you're describing only happens in Condorcet systems. IRV (Instant Runoff Voting) wouldn't lead to that, as the "third place guy" would get eliminated in the first round
Condorcet FTW!
The description above is _exactly_ why I prefer IRV. It forces candidates to run on their own merits rather than just attack each other. You actually need significant first-place votes to win
I think Condorcet would sufficiently incentivize positive messaging, since the cost of attack ads is borne by one candidate but the benefit is shared with all the other non-attacked candidates.
The Alberta election seems to have been the result of the campaigns not understanding the incentives, which I would expect to sort itself out in a cycle or two.
Stelmach may objectively suck, but so long as majorities prefer his nonsense to either of the other candidate's he's still the least-bad choice. And without a motivated base of support, his worst policies will likely be comparatively easy to unwind.
IRV is worse at picking the right winner who maximizes voter satisfaction.
https://electionscience.github.io/vse-sim/vse-graph.html
it has tons of other problems, like entrenching duopoly.
https://electionscience.github.io/vse-sim/vse-graph.html
and pretty much any advanced voting method encourages more civility, in order to not lose support from supporters of your philosophical allies.
no. cardinal methods (score voting and approval voting) are probably better in real life or at least nearly as good, and vastly simpler, cheaper, more politically viable, and better for escaping duopoly.
http://scorevoting.net/AppCW
https://electionscience.github.io/vse-sim/vse-graph.html
https://asitoughttobemagazine.com/2010/07/18/score-voting/
I think in an ideal world, one of the more complicated voting methods would be ideal… but approval voting seems to me the perfect combo of likely to have an impact, easy to understand and implement, and above all, the absolute least likely to generate voter’s regret! Because if any of your approved candidates win, you’re “happy” with your vote. It does exactly what it says on the tin. Want to protest vote? You can. Like a major party person and also a third party person? Do it. At least on an individual level, it’s empowering. And I think it does enough to at least crack the door to third parties and less polarizing politics.
By contrast I attempted to explain ranked choice voting to my relatively smart parents who actually were in a spot in Oregon trying it out, and basically ended up saying something to the tune of “just rank like they tell you to and trust that it works magically”. But that’s pretty unsatisfying.
Of course if you think moderation and civility are bad, approval voting is not for you. But I think most people like those things?
approval voting is 95% as accurate as the more complicated voting methods and radically more politically viable. seems to me about 20-100 times more impactful than everything else here, or really anything else the EA community is talking about, from a cost-benefit POV.
https://electionscience.github.io/vse-sim/vse-graph.html
we also need election by jury.
https://www.electionbyjury.org/manifesto
approval voting is 95% as accurate as the more complicated voting methods and radically more politically viable.
https://electionscience.github.io/vse-sim/vse-graph.html
> Someone (I think it might be Paul Graham) once said that they were always surprised how quickly destined-to-be-successful startup founders responded to emails - sometimes within a single-digit number of minutes regardless of time of day. I used to think of this as mysterious - some sort of psychological trait? Working with these grants has made me think of it as just a straightforward fact of life: some people operate an order of magnitude faster than others.
I know a few people who respond to emails quickly but not very well, and I also feel like this sort of “respond to emails quickly” mentality lines up well with people who don’t take their work-life balance very seriously either—which is probably a good trait to have if you want to be a startup founder.
I’ve been through periods where I respond very quickly and periods where I need reminders to respond. It is absolutely a window into mindset, energy, motivation, dedication and prioritization.
Someone who responds quickly is on top of things. Someone who doesn’t is mentally overwhelmed, distracted, disorganized or depressed.
Very impressive to read all the good things that came out of that, thanks to everyone involved for making the world better.
Isn't IRV strictly better than Approval Voting? This just seems like a bigger win than expected.
Noooo, don't you dare start this again!
Again?
Where can I find the previous start of this?
The case for ranked-choice over approval: https://fairvote.org/resources/electoral-systems/ranked_choice_voting_vs_approval_voting/
The case for approval over ranked-choice: https://electionscience.org/education/approval-voting-vs-rcv
I once attended an online debate (hosted by an EA group) between representatives of these two orgs, but I don't think it's online anywhere.
See also https://xkcd.com/1844/.
For reference, the case for instant runoff voting, from that link:
1. A lot of places already use IRV. Approval voting is new, and dangerous.
2. In two case studies that we characterize as using approval voting, of which one actually does use approval voting, people didn't vote for enough candidates. This pathological behavior comes from the fact that approval voting rewards you for restricting your votes; additional votes dilute the power of each vote.
3. However, white voters voted for more candidates than black voters did, giving whites a larger say in the election than blacks got.
4. 51% of voters (in approval elections) were able to correctly describe how it worked.
5. No comment on how many IRV voters were able to correctly describe how it worked.
6. Dartmouth experimented with approval voting, but abandoned it after several student presidents were elected with support from less than 40% of voters.
7. No mention of the fact that the replacement was plurality voting ( https://students.dartmouth.edu/student-life/sites/students_student_life.prod/files/student_life/wysiwyg/spring_2025_election_code_pdf.pdf , clause 6.7.2 ), which doesn't even pretend to address that "problem".
8. IRV "improves representation for women and people of color".
approval voting is radically superior to IRV and vastly more viable from a cost-benefit standpoint. it passed in fargo and st louis by 64% and 68% respectively.
> In two case studies that we characterize as using approval voting, of which one actually does use approval voting, people didn't vote for enough candidates.
there is no such thing as "enough" candidates. it's perfectly possible to elect the social utility optimizer with everyone approving only one candidate. approval voting has performed fantastically in multiple simulations which incorporate varying percentages of tactical "psychopath" voters, who absolutely "bullet vote" for only one candidate whenever it's the best strategy. so this is not a real issue, it's just a misunderstanding of basic math. lots of discussion by math phd's and other assorted experts here.
https://voting-in-the-abstract.medium.com/voter-satisfaction-efficiency-many-many-results-ad66ffa87c9e
https://www.rangevoting.org/BulletVoting
> additional votes dilute the power of each vote.
> white voters voted for more candidates than black voters did, giving whites a larger say in the election than blacks got.
you just made these two back-to-back contradictory statements.
the strength of your vote is not strictly a function of how many candidates you approve.
https://www.rangevoting.org/RVstrat6
> 6. Dartmouth experimented with approval voting, but abandoned it after several student presidents were elected with support from less than 40% of voters.
this has been debunked to death. how have you not done basic research on it?
https://www.rangevoting.org/DartmouthBack
> IRV "improves representation for women and people of color".
there's extremely limited evidence on this, and every reason to believe approval voting would have a greater effect.
https://electionscience.org/education/st-louis-success
https://www.rangevoting.org/IRVraceMinorities.html
your entire response is a stream of admissions that you haven't seriously researched this topic.
Maybe you should try reading my response a little more... intelligently?
I didn't say those were my ideas. I said "the case for instant runoff voting, from that link". Those are the points presented in the link Taymon A. Beal provided. In your model of what my comment says, what is the meaning of points (5) and (7)?
As you note, fairvote's "arguments" are stupid and explicitly self-contradictory. I felt this was obvious enough that there was no need to call it out.
> there is no such thing as "enough" candidates. it's perfectly possible to elect the social utility optimizer with everyone approving only one candidate.
On the other hand, that one is easily fairvote's strongest argument. If everyone approves only one candidate, you're doing nothing other than using plurality voting. Reforming plurality voting into more plurality voting is not a worthwhile use of time.
fairvote are professional liars, and also just don't understand voting theory. literally everything they've published on this has been debunked.
https://electionscience.org/library/approval-voting-versus-irv/
There are many drawbacks to IRV/RCV that aren't present in approval voting. Here are four I detail in an old post of mine:
* Ranked-Choice Voting Encourages Strategic Voting
* Ranked-Choice Voting Doesn't Always Result in the Favored Head-to-Head Candidate
* Ranked-Choice Voting Allows for Election Spoilers
* Ranked-Choice Voting Allows for Candidates to Receive Higher-Ranked Votes but Have Worse Outcomes
https://thegreymatter.substack.com/p/the-case-against-ranked-choice-voting-part-i
IRV is a strict subset of RCV.
You are attributing to RCV issues that are specific to IRV.
Condorcet RCV dominates both IRV & approval.
Yeah, I was using the common usage of RCV, which is interchangeable with IRV. I should have been more precise.
IRV and RCV are used as synonyms for the most part nowadays.
no, approval voting dominates condorcet, in most realistic circumstances involving strategic voting. approval voting always elects the condorcet winner under realistic models of strategic voter behavior, whereas tactical condorcet voters can paradoxically fail to elect the condorcet winner. this is intro level voting theory.
http://scorevoting.net/AppCW
http://scorevoting.net/BayRegsFig
http://scorevoting.net/CondBurial
nevermind the dramatic superiority of approval voting in terms of simplicity and political viability.
I think that most of those points only work in pretty contrived scenarios, though, and in realistic elections with realistic preference flows and voters who aren't time-travelling wizards they don't really come up.
Did you see the Part II post (https://thegreymatter.substack.com/p/case-against-rcv2) that I linked to at the bottom, where I show a real-world situation? It shows that there was an election spoiler (Wright), the best head-to-head candidate (Montroll) didn't win, some voters would have been better off had they been strategic (Wright voters voting for Montroll instead), and even the weird case where a candidate (Kiss) does better but then loses the election.
Of course, voters aren't going to know exactly how to be strategic to get the outcome they want, but isn't the fact that it's possible a demerit for the election system? Is the fact that I found all of these in one of the few elections of this type somewhat convincing? If not, what do you think would convince you?
no! approval voting is radically superior.
https://electionscience.org/library/approval-voting-versus-irv/
The rule of thumb for selecting your cutoff in approval voting is to approve every candidate you like more than the one you expect to finish second. Even the rule of thumb is very complicated to enact because it relies on potentially unavailable information.
I guess two front runners would probably be common even after approval voting is rolled out for some time, because of psychology and limited attention. So most people could just choose every candidate they prefer over the leading candidate on the opposing side (the one they get the most ads for, say).
But if there is no front runner of the opposing ideology, you'd need a prediction market to be in place before you could figure out how to correctly cast your vote. If you don't fit into the two standard sides, it's especially difficult, since you wouldn't have a way of guessing which of the two frontrunners to even consider or use as the benchmark for your ballot.
There's a lot to like here, the anti-mosquito drones in particular look like they'd have both public-health utility and market potential. The AI safety advocacy probably dwarfs anything else going on here in terms of magnitude, it's just hard to know what actions are actually going to be relevant so the existence of advocacy groups that can pivot as the industry changes is really the only way to handle that. (I don't have any background in grants, and am terrible at writing and funding them, and have no worthwhile opinion on whether you should keep doing this, but I do think you continue in some fashion to try to place people in key policymaking spots ahead of the big AI decision points.)
I'm somewhat baffled that the animal welfare team convinced farmers to selectively abort all the male chickens. I eat a bunch of chickens, and eggs, so this is "not my fight" I suppose. I understand that under some value systems this makes sense, but it definitely approaches a danger zone, and I gather there has been some heated discussion of "negative utilitarianism" lately. Chickens don't have future-oriented mental states, perhaps they don't even have preferences at all aside from avoiding pain, in which case this approach could be correct.
I saw a reddit discussion about this and it was suggested that farmers could raise the male chickens for meat, and just don't because they'd have a "surplus" of meat they couldn't sell which ends up going to developing nations, and this outcome was described as being bad because that weakened the local chicken farming economy in those developing nations. That seemed like a rather questionable justification to me, perhaps that ought to be looked into. If I were an activist against factory farming, from a practical purpose I don't know if I'd want to lose "dude they put all the male chickens into something called a grinder a day after they're born" as an argument by just killing them all in-ovo, they just sane-washed factory farming for an outcome that may amount to the same thing.
I'm confused about your objection; even if chickens have subjective experiences, chicken *embryos* probably don't, so I don't see anything wrong with selectively aborting them?
Agreed on the mosquito drones as a visibly interesting approach to a perennial problem. Though…maybe one with some unfortunate consequences. I get that humans don’t output a specific ultrasonic frequency, and aren’t quite as vulnerable to a small electric charge…
Regarding male chickens: I don’t see how aborting male chickens could be any *worse* than the status quo. Even if we assume the zygote gets exactly the same internal experience as the chick would, it just happens sooner.
What exactly is the “danger zone” you have in mind?
The formulation by which it’s a good act to abort them rather then let them be born and live a day and then be killed is one that can lead to bizarre calculations of utility. In very fact-specific individuated cases it may be fine to argue “this person or creature has so much suffering it would have been better to prevent its existence” but I don’t think I agree with it even then, you have to invent pretty extreme hypos for it to make intuitive sense. So I’m suspicious of applying that kind of formulation to an entire class of anything this broad. The one day of life probably isn’t trivial, chickens presumably don’t have a will or self/concept but they are programmed to do a few things, and in that one day they would bust out of their egg, look around, imprint on a nearby creature, maybe that’s a really good chicken-day. “One day of life before you’re tossed into the grinder” even kinda sounds like one of those extreme hypotheticals, and it still seems dubious that nonexistence would be better than existence there.
I would assume that they would lose money raising the male chicks for meat because they would be competing with meat from chickens bred to optimize meat production and wouldn't be able to sell the rooster meat for enough money to pay for the cost of raising them.
Hey CJW! Just wanted to provide some context that may help understand the situation a little better:
- On selling the meat, you're liking referring to something that happened back in the late 2010s. After Germany and France banned the practice of male chick culling, in-ovo sexing technology was still fairly rudimentary at that stage, and not all producers had access to it. As a result, they reared the males and sometimes exported the meat to other countries. The reason it was cheap was because it was a byproduct that they had to get rid of, not because doing this is particularly economical. The other commenters who have pointed out that males of the laying-hen breeds are uneconomical to raise for meat compared to commercial meat chickens are correct. As in-ovo sexing technology further developed pretty much all producers chose to use it rather than raise males for economic reasons. In the US, it's not possible to slaughter significant numbers of males of the layer breed because they're a different size and shape than the typical meat birds we eat, so slaughter facility infrastructure isn't set up for it. Even if it was, it wouldn't be economical, and many people report not liking the quality of the meat anyways since it tends to be tougher than the meat we're used to eating.
- On whether in-ovo sexing amounts to the same thing as maceration, there have actually been a few studies on the cognitive development of chick embryos over the course of the 21 day incubation period. The evidence could always be a better, and ultimately consciousness isn't something I think we fundamentally understand, but generally in-ovo sexing works before there's meaningful evidence that the embryos are conscious or feel pain.
- Activists against factory farming can and have made the argument that you describe, but Innovate Animal Ag isn't that class. We're activists for better animal welfare. Welfare is low enough during day 1 inside a hatchery that it's a clear win in my mind to not experience it, under many ethical theories, not just negative utilitarianism.
Hi, thanks for the reply, I believe that does explain what I had read because it concerned an EU poultry market.
I also find it amazing that the cognitive development of chick embryos has been studied. I still lean towards thinking "existence at all must be better than non-existence even in this case", though there is likely some point in development before which I probably wouldn't consider it the "same" individual, so that could be relevant. If there were a process that made eggs never fertilize with male chicks in the first place, like if they put the roosters' sperm in a centrifuge or something, that would seem better to me if the price could be made comparable. As I said, I do eat chickens so I don't exactly have a moral high ground here, but I guess I still care a nonzero amount. Anyhow thanks for the explanation.
On Grant #14 on the list, as second author on the paper (as an advisor on the project that contributed, but was not paid by the grant,) I strongly second that it cannot be evaluated on its own. That's the problem with marginal work promoting changes to complex systems; even if we get HCTs for the next pandemic, or get more of them for existing diseases, you'd need to do an infeasibly complex Shapley-value analysis across tons of efforts and projects with tons of unknowns to attribute part of the impact to this grant.
But, critically, difficulty measuring impact doesn't function as much evidence about extent of impact, and I think that the paper helps shift the burden of evidence for future studies and thereby saves significantly more lives in expectation than the cost. (I'm very comfortable with that estimate, albeit with low confidence about magnitude.) And I'm grateful that grants like this can be made by funders who can appreciate that it's hard to measure impact, and the expected impact is still the critical factor by which to judge grants - even though subjective estimation of that counterfactual impact remains unavoidable.
Regarding RadVac; I initially read Wentsworth's post on LessWrong about "Making Vaccine".
I went ahead did the same, and administered it to ~40 people, none of whom died of Covid (although in expectation none would have anyway). I also took their 40-page whitepaper and made it into a 2-page policy brief, and created an easy-to-follow instruction page for administration.
I also got involved with the RadVac team, and helped them get their proposal in front of high-ranknig government officials in several countries; ultimately none were willing to sponsor challenge trials, but we knew going in it would be a high-variance / high-expectation long-shot.
I think there's something to be said here about "pulling sideways". It was definitely a lot of effort, but there were some relatively-close almost-successes. It has significantly improved my belief that it's possible to make an impact in labor-limited areas.
> "This one is confusing to evaluate; the specific proposal failed, it encouraged its opponents to create a distraction proposal to sabotage it, and the distraction proposal unexpectedly passed, meaning that Seattle did get a more interesting voting method after all (although unclear whether it’s good). Is this a success of our grant?"
Fun fact: this is also how sex discrimination was banned in the United States (poison pill added to the Civil Rights Act of 1964 that failed to peel off enough votes to kill it), so I think you're in good shape calling this a success.
I read an interesting article about this recently which I can't find right now, but the story is actually a little more complicated. Fortunately Wikipedia has the basics of it: https://en.wikipedia.org/wiki/Civil_Rights_Act_of_1964#Aspects
In short: Howard Smith, who got that part added, may have intended it as a poison pill; but the National Woman's Party, who encouraged him to add it, definitely intended it sincerely (even if perhaps they thought the reason he'd add it was as a poison pill).
(Although, now that I read this Wikipedia article, there seems to be more reason than I thought to believe that it was in fact always meant sincerely and not as a poison pill.)
I stand informed!
Hey! Luca De Leo here from the 2021 round. I don't think I ever got an email asking for an update but I just filled in the form at the top.
Thanks a lot for everything you do!
Im the founder of Spartacus.app, the conditional commitment platform and 2024 grantee.
Just wanted to thank Scott again for the help in getting the project off the ground.
We still have an opening for a summer intern!
Dm me if interested.
Regarding #6, I was that climate activist who wrote a rock opera, but it really wasn't a climate activism theme. More a mytho-poetic, sci-fi, rumination on love with some inclusion of terraforming snuck into to the plotline. I do like the idea of using music and theater to capture the public's attention to pressing topics like climate change, and truly do appreciate the link to my opera recording.
61 certainly sounds like something I'd participate in. I seem to be unusually good at learning across a wide variety of areas, so I might provide some unusual data.
This is fantastic! It's inspiring to get this update about ACX grant's support of rational progress. Thank you.
Hello - founder of Innovate Animal Ag here! Responding in particular to this: "One thing I still don't understand is that Innovate Animal Ag seemed to genuinely need more funding despite being legibly great and high status - does this screen off a theoretical objection that they don't provide ACX Grants with as much counterfactual impact?"
First of all, thank you so much for your support, and for your kind words. ACX has been extremely helpful to us in many ways including to but not limited to funding.
To you, we may seem "legibly great," but many people don't share your perspective. We have a very different approach from most animal welfare organizations in that we work very collaboratively with the animal ag industry to find win-wins - technologies that are good for welfare but also have a value proposition to farmers. Many from the traditional animal advocacy world are skeptical of this collaborative stance, and some even find it heretical. This, combined with the fact that most donors are still mostly interested in companion animals rather than farm animals, means that ~90% of donors who traditionally fund animal welfare wouldn't fund us.
In other words, it's great to have donors who are weird in similar ways that we are weird. It's possible that we could have raised this money from other sources, but this would have taken a lot longer, and distracted a lot of focus from program work. It also may have pushed in directions that are more legible to traditional funders, but ultimately less impactful.
Another counterfactual benefit of the ACX grants is that we hired multiple people either from ACX classifieds, or who read about us through ACX grants. Again, the reason for this is because your audience is similarly weird to us, and is particularly likely to be excited about our unique theory of change.
+1 for finding innovative win-wins
My jaw dropped when reading the review, from my perspective you've been astonishingly successful. A truly enormous amount of good done for the money, sincere congratulations.
Thank you! 🙏🏻
Are you doing work in the UK as well, or focusing on the US?
We're primarily US focused, but may do a small amount of work in the UK in the future.
As a Seattle voter who wanted approval voting and is also uncertain about whether IRV is a net improvement for this specific circumstance: I consider this a successful ACX grant. One of our biggest weaknesses as a civilization right now is status quo bias and specifically a strong bias against experimentation and trying new things. This resulted in a major municipality trying a new thing, which is Good.
Labelling of Endocrine Disruptors:
Will this change your choice in Beer? If so, why or why not? If it will, how confident are you that anyone else will also change? (If so, you're going to have to overturn a lot of the craft beer lobby in order to get your labelling implemented.)
Unless you only mean to label endocrine disruptors in shampoo? (Which is a very underlabeled product).
Getting instant runoff voting to pass is absolutely a success. Instant runoff is better anyway (Sorry to the approval voting partisans, but they're the one with the fringe view here.) It's probably a low investment lift to get ranked ballots to pass in other places, particularly rural parts of the west coast.
> (Sorry to the approval voting partisans, but they're the one with the fringe view here.)
That's not really an argument, that's pure argument by popularity.
Personally, I think of IRV as being a lot like Esperanto; it's the first alternative that people tend to encounter, and therefore is championed by those who have realized there's a problem but *haven't* realized that there might be more than one approach to a solution and who therefore have stuck to the first one they encountered. Which isn't saying you're wrong, necessarily (although I think you are) -- just that there's an explanation here for popularity here other than it being the best, just as Esperanto isn't the best interlang (on which subject I must of course link to http://jbr.me.uk/ranto/ ).
(Yeah OK I realize that the international language problem that Esperanto is trying to solve is pretty fake because in reality people just use English, but, the comparison still holds IMO :P )
The full argument is a bit involved, but basically approval voting is vastly more subject to gaming than ranked ballots are. The details are technical and beyond what I can explain in a blog comment and in the past when I tried to explain it to approval voting enthusiasts it proved to be a waste of time because they didn't understand the issues and weren't interested in learning.
Sure, that's the start of an actual argument! I do have to say it sounds unlikely offhand, but whatever, that's not the point right now. :P
this could not be more wrong. approval voting is incredibly resistant to tactical voting. see bayesian regret figures from princeton math phd warren smith and voter satisfaction efficiency metrics from harvard stats phd jameson quinn.
http://scorevoting.net/BayRegsFig
https://electionscience.github.io/vse-sim/vse-graph.html
there's even a mathematicaly PROOF that approval voting elects the condorcet winner (a pretty good result) even if 100% of voters are maximally strategic.
https://www.rangevoting.org/AppCW
i would really not preach to people about not being interested in learning, when you could have learned this with 10 seconds of google searching and 5 minutes of reading.
You fundamentally misunderstand that result. It says that if votes are maximally strategic - a difficult thing to do in practice and a very unrealistic assumption - then it will get the condorcet winner. This is in contrast to what happens if they aren't strategic, which is they won't, because of obvious reasons like voters who disapprove of both of the leading candidates having their votes not counted. Requiring voters be strategic is a bad thing. What always gets the condorcet winner regardless of how strategic or non-strategic voters are is ranked choice ballots using a condorcet algorithm.
first of all, i co-founded the center for election science with warren smith, the princeton math phd who created that site. i've contributed to several pages on the site, going back to 2006—including this one on...optimal tactical score/approval voting:
https://www.rangevoting.org/RVstrat6
so i know what the page says.
you are demonstrably mistaken about several core voting theory concepts.
misunderstanding: you seem to think it's desirable to elect the condorcet winner. this is false. it's _mathematically proven_ that an electorate may prefer someone else, possibly even the condorcet _loser_. this is one of the most elementary points of the field of voting theory, and largely why arrow's theorem is noteworthy. it's explained here in layman-friendly terms.
https://clayshentrup.medium.com/a-simple-proof-that-majoritarianism-is-wrong-5ac15b195b66
the CORRECT measure of voting method performance is "voter satisfaction efficiency" aka "bayesian regret". score voting and approval voting outperform condorcet methods in a wide range of scenarios.
https://www.rangevoting.org/UniqBest
misunderstanding: What always gets the condorcet winner regardless of how strategic or non-strategic voters are is ranked choice ballots using a condorcet algorithm.
utterly false. this is one of the most common novice misunderstandings in the field.
https://www.rangevoting.org/CondBurial
and since most users of ranked voting methods will just assume it works like borda (literally what most people i've spoken with in san francisco and berkeley when i lived there thought), they'll just naively strategically exaggerate anyway. all of this is common knowledge to those of us who've worked in this field for more than a week.
incidentally, you can see computer simulations here showing that, under plausible models of voter strategy, approval voting is BETTER at electing the condorcet winner than actual condorcet methods.
https://www.rangevoting.org/StratHonMix
misunderstanding: "Requiring voters [be strategic is a bad thing."
i'm not saying we REQUIRE them to be strategic to get a good result. i'm saying that it's good to have a fairly benign, even GOOD, "worst case scenario" in the event that many of them ARE strategic. because in the real world, people WILL be strategic.
if voters are honest under approval voting, they can do BETTER than getting the condorcet winner, because in many cases, the social utility maximizer won't be the condorcet winner.
Approval voting is not a loophole to Arrow's Theorem. It has the same limitations as ranked choice ballots. I'm done discussing this with you because you have once again demonstrated that you're a crank and interacting with you is a waste of my time.
i learned esperanto many years before i co-founded a non-profit to advance approval voting. :D
all the evidence says approval voting is better than instant runoff voting. like in literally every way. more accurate, simpler, precinct summable, cheaper, etc.
https://electionscience.org/library/approval-voting-versus-irv/
“I’d be interested in hearing from research psychologists about whether this project [clearerthinking.org] is well-known in your field, and how effective it actually is about keeping you guys honest.”
I’m a research psychologist and never heard of it. I love the term “importance hacking”. I think that’s a very real phenomenon, but I think they’re missing a dynamic that’s causing it.
The most prestigious journals are the most generalist (eg. Science, Nature). Because these journals get submissions from every possible subfield, their editors are necessarily less expert in each. They judge which papers to send out to review based on how important the paper seems, but they lack the expertise to know, opening the door for hype to be effective. They also don’t know the relevant experts to ask to review and often ask high-profile researchers in the bigger field who are not actually expert in the relevant subfield, and often don’t know that something isn’t entirely original or as important as how it’s being hyped. Some researchers have figured out that if they hype their study well (often not just the result, but also the importance of the question being tested), they’ll actually have an easier review process AND a higher impact publication. Most researchers I know quietly agree that the most important publications in their area are in somewhat more specialist journals, where the editors are area experts and the reviews are tougher.
A feedback mechanism that could counteract this is that those hyped papers should ultimately get cited less by the experts who work most in that subfield, since they’re not actually that important. But publication in high prestige venues tends to break that mechanism. For one, the papers are highly visible and end up being cited outside the subfield by other non-experts who don’t know better either. For another, there’s usually only a couple papers per year published in Science in a given subfield and everyone in that field knows those papers. Both expert and non-expert reviewers will expect you to cite them.
One solution would be to have more editors who are expert in more areas at these journals. But that costs the journals money (unlike reviewers, editors are paid, and at these journals they’re full time editors, not working scientists) and there’s currently no incentive for the journals to do it.
24, per their own description, is trying to make animal agriculture more productive in ways that align with cost efficiency. I don't know that much about bio stuff so am not sure whether they can, but if true, that further entrenches animal agriculture and diminishes the market share for plant-based products. This seems like it's extremely bad from an animal rights perspective, which it appears that you take, and would probably make that grant net negative if there's a sizable chance they actually do anything.
Animal agriculture is often conducted on land that is impractical for plant-based farming. Is your thought that cattle should be driven extinct, or let roam free and wild? (Animal rights cattle raising is to feed them grass, let them live outside -- with a leanto for shelter when it's horrid outside, and otherwise make sure they get pretty happy lives, doing what they'd do in the wild. Then they have one bad day. This is not very different from hunting, if you aren't hunting white-tailed deer (which are destroying our forests).)
I'm not advocating for any specific view (sorry, was logged into my old account originally), just pointing out that by Mr. Alexander's lights in this comment (https://www.astralcodexten.com/p/acx-grants-results-2024/comment/49227604) it should be considered harmful.
24 was originally working on IVG for humans, and we funded them on that basis. They pivoted to cows years later. I suspect the cow work is just a testing ground before they move back to humans. If the cow work does speed up IVG for humans, I'll accept it as a necessary cost.
Thanks! Personally I'm skeptical of this but I do understand where you are coming from.
You can donate to ACX Grants here: https://manifund.org/acx-grants
I had the mistaken impression that this was a closed fund for Scott and his immediate network, but it looks like anyone can chip in.
(It looks like that page could use updating – it looks like it was last edited shortly after the 2024 round closed)
That Manifund page should continue to work, but fwiw, if you're interested in donating to ACX Grants, I'd suggest holding off for now, and wait for Scott to put out the "lingering questions & next steps" post!
Congratulations to Scott and all the grantees!!
>44: Advocate For Pandemic Response Team At FDA
>This team prefers has asked me not to discuss their progress publicly, but you can probably guess what their lives are like right now, and your guess would be correct.
Dumb question by a non american. Are they working on the current pandemic or fired by Trump/DOGE?
If I were to guess (as an American) I would say they are experiencing the non-optimal timeline : -(
Just to say it out loud: This is an astonishing success rate given the very light process (which was the point), congratulations to the grantees, but also Scott and the reviewers.
Iny view, the entire spending is easily justified by 1-2 of the successes, such as Innovate Animal Ag, and there's potential for dramatically more (such as maternal care and lead batteries in Nigeria).
I weight the AI safety stuff at near 0, but that doesn't matter at all here - even aside from the big real world successes, there's a surprising amount of real research and development done here.
If the mosquito drone guys ever want to expand operations, they should look into making anti-invasive species drones for the New Zealand government. The NZ government has a "plan" to be rat-free by 2050 (I think that's the year) but it will require new tech.
> 62: Advocate Repeal Of Interstate Runaway Compact
Could someone explain why this is a good idea and what it would improve?
It deals with how easily police can force runaway teens to return to potentially abusive families who they don't want to return to. See discussion at https://www.astralcodexten.com/p/acx-grants-results-2024/comment/49237506 .
> 4: Alice Evans’ Research On “The Great Gender Divergence”
> Meanwhile, Latin America and the Caribbean face a different obstacle: pervasive violence elevates femicides.
Androcides outnumber femicides almost everywhere in the world; in no country in Latin America or the Caribbean are women more than 35% of all homicide victims (40% if we include Guyana, which is South American but not Latin). https://en.wikipedia.org/wiki/Homicide_statistics_by_gender
Sure, men also kill men, but that does not help women who are killed by men in any way
It doesn't, but (as I thought was obvious) my point is an increase in femicides (is that what "elevates" means in this context?) isn't gender inequality/divergence, which was the context.
> the distraction proposal unexpectedly passed, meaning that Seattle did get a more interesting voting method after all (although unclear whether it’s good). Is this a “success” of our grant?
I think not for two reasons. The first is that IRV has sufficiently different properties than AV that I think the two voting systems achieve fundamentally different goals. (AV avoids center-squeeze, IRV does not. AV has a lot of nice monotonicity properties, IRV does not, etc.). The second reason is more of a quibble, which is that I think more needs to go into entrenching whatever voting system is established, whether that be legally or by means of raising public support. IRV systems are frequently repealed (sometimes by voters, whereas, in the US, AV has only been repealed at the state level in spite of voter support) even after initiatives pass, and I'm not sure the Seattle case should be expected to last.
Hello, I work on the frontend for Baby Steps (#60, helping prospective parents choose IVF clinics with a success rate prediction model). Thanks so much for the grant! Sorry I didn't thank you in person while I was in Berkeley.
If anyone has any questions about it, I can answer them in this thread :)
Thank you, Scott! Without you or the grant competition, we would never have built Manifold.
Even the idea of user-created prediction markets was something we thought of a few months before ACX grants while reading your Mantic Monday series.
I knew I wanted to work with Austin, having met him online through Discord. And the grant competition was the first thing we decided to collaborate on. We brainstormed ideas and landed on user-created prediction markets (on Solana haha).
Austin flew to meet my brother and I, and the next day was coincidentally our final round interview for YC. We pitched what would become Manifold in a Zoom call on our living room couch. It was basically the first time the three of us ever met. They didn't accept us though, haha. They didn't like our two-week old crypto idea, but that didn't slow us down.
It was December 2nd. We worked every day until Christmas to build the first version of Manifold. After a call with Zvi Mowshowitz, who was very negative on crypto, we came up with the pure play money idea, and built it out.
It was ready the day Scott emailed us to say we won. Scott posted the link in the ACX winners post; we instantly got hundreds of users. The rest is history.
This is Alene from #25 (Legal Impact for Chickens).
So grateful to ACX and all the ACX readers who have supported LIC!
Also since we sent in our update, we have had a new victory: LIC proved in court that SPCAs can sue to stop cruelty to farmed animals. California's Humboldt County Superior Court just ruled that LIC's cruelty lawsuit can proceed against well-known dairy operation Alexandre Family Farm, which supplies Whole Foods.
California has a law against animal cruelty—which extends to farmed animals. But the fact that prosecutors rarely enforce cruelty laws against the powerful corporate interests that control animal agriculture leads some people to falsely believe that animal ag is above the law.
The court's order determined that LIC's cause of action, enforcing the cruelty law against an agricultural operation, is valid. The court ruled that an organization like LIC can directly sue ag companies for cruelty, without needing to pass any additional bar for standing.
We're excited to move forward with the facts stage of the case. And we hope other California-based organizations will start to bring other cases like this for farmed animals, too.
Congrats, and I'm glad about the work you're doing! If you spot any ways to force companies to try harder at preventing coccidiosis and salmonella infections I think that could have a very large impact, because I recently took in some rescue hens with coccidiosis and the improvement in their happiness once they were treated for it was quite large. Also if these diseases are prevented rather than managed it may slow down the development of drug resistance.
These are great points. Thank you so much, Catmint. And thank you for rescuing birds!!! The last USDA administration did put out a call for comments about preventing salmonella from getting into the food people eat. And LIC commented asking for solutions that would improve animal welfare, and prevent birds from getting salmonella in the first place. https://www.legalimpactforchickens.org/salmonella . But sadly, the current administration withdrew the proposed rule. :-( Still, there are probably solutions that don't involve the government at all . . . Like maybe individual farmers could simply choose to vaccinate birds for coccidiosis and salmonella? I don't know anything about whether farmers already do that, or whether they could. But I will ask around.
Vaccines would need to be used carefully, because the only scalable options we have right now (as far as I can tell from my brief research) are live vaccines, and those come with a risk of introducing the disease to a flock or a building that did not already have it. But they're definitely very useful for places that do, which unfortunately is most of them. Probiotics apparently help, but I'm not sure how much. There's also some bacteriophages that attack salmonella, which sounds very promising, but as far as I can tell this is still an active area of research rather than a commercially available product. (This talks about both: https://pmc.ncbi.nlm.nih.gov/articles/PMC10672927/) LIC's suggestions of decreased flock size and earlier testing also sound effective. Glad you guys already looked into it.
Mmm Thank you for all the detailed info.
Hi Scott! Longtime fan of your writing here. As it matches this ACX Grants article's topics of "climate activism without an associated musical theater component," and " lobbying/advocacy organizations," I'd love to share what I'm working on.
Since 2024, in addition to my personal Substack The Weekly Anthropocene (https://sammatey.substack.com/, thanks again for linking to my interview with Ozy Brennan!), I've been writing Your Dose of Climate Hope (https://climateactapp.substack.com/), a Substack partnered with the new Climate Action Now advocacy software startup that's providing daily actions to advance hopeful, sensible, incrementalist, life-improving, innovation-incentivizing, pro-permitting reform, and pro-Abundance Agenda climate policy. In many cases, I'm pretty sure that Your Dose of Climate Hope is the only advocacy entity trying to contact federal, state, or city governments on a particular issue, especially our regular feature of highlighting a promising emerging technology (e.g. advanced geothermal, modular housing, AI-enhanced crop breeding, one-dose malaria vaccines, ocean alkalinity enhancement, and much more) and asking policymakers to ensure a supportive legal, regulatory, and permitting process. I think in many cases our letters are going to be the first that a given politician has ever heard of the topic, and we're trying to get in there early with a first impression to show that constituents want to build this innovation.
If you find this interesting, I'm not looking for money, but a Substack Recommendation from Astral Codex Ten would be INCREDIBLY valuable - the Recommendation function has been the primary way I've grown my audience for both projects!
I would posit that a Substack Recommendation amounts to ~2 minutes of your time to directly help advance the epistemic state of climate activism and indirectly help positively shift U.S. state and federal policies.
20: Microbes From Beetles That Can Digest Plastics
24: In Vitro Gametogenesis Startup
Both of these two seem like some good science got done, which might have results way down the line but not any time soon
32: A Robotic System For Automating Cell Culture Media Testing
This sounds kind of like what Promega does?
Hi! This is Michael Todhunter, the guy from #32. Promega indeed offers robotic lab hardware - as does Hamilton, Tecan, Beckman Coulter, and a bunch of other companies. What makes this grant proposal different is that I'm trying to make, for lack of a better term, a self-driving biologist - a robotic system that can plan, execute, and interpret its own experiments. We might buy stuff from Promega, but we're not competing with Promega. FutureHouse is working on similar projects, but I'm focusing on embodying an automated scientist in a robotic lab, whereas, to the best of my knowledge, they're focusing on LLM-based scientific reasoning.
impressed! this gives me reason to give ACX my monthly dollars.
The mosquito drones sound very exciting.
They said they expect to be killing mosquitoes by June. Have they succeeded? If not, what's the latest estimate?
We've had a few setbacks but we're getting to it. Expect it in the next few months!
> I'm most ambivalent about our science grants. Many of them say they are successful and can point to published papers which explain the science they did. But it's hard to judge whether anything useful has changed based on the science getting done. I know it's important to fund basic research and not just last-mile technology startups, but it's hard for a mini-grants program like this one to evaluate these kinds of abstract interventions.
I can think of a couple of reasons for this.
1) Science is just really hard *and* really expensive. A single (reasonable) publication can costs hundreds of thousands of dollars to produce. Even just one grad student can cost $75-100k per year when all costs are factored in. Plus we already have systems in place to review and fund science. They aren't perfect, but it usually means there's unlikely to be any low-hanging fruit to be picked unless perhaps if you have a very niche interest.
2) Grant proposals are usually reviewed by experts in the field with many years (often decades) of experience. If a person submits a proposal to you, it likely means they are unable (or expect to be unable) to get it funded through the normal channels and are hoping you and your team will be less skeptical than experienced, field-specific reviewers. This doesn't mean the idea is bad per se, but at least it may not be fully formed yet.
What some foundations do is fund projects at an early stage in order to generate data for proposals to federal agencies. They will make a second year's funding contingent on submission of an NIH proposal, for example. This seems like a better way to leverage a small amount of money to have a bigger impact.
We interviewed Kenneth Arrow in 2012 and he acknowledged his theorem doesn't apply to cardinal voting methods.
https://www.rangevoting.org/ArrowEndorse.html
It was also covered in a recent veritasium episode.
https://youtu.be/qf7ws2DF-zk?si=A6ggrAP7Sc02Fss5
There's also a mathematical proof of this here.
https://www.rangevoting.org/ArrowThm
I have no idea why you're interested in the Dunning-Kruger approach.
> They’re making some progress on a salt reduction campaign in a developing country they prefer not to name publicly.
I'm concerned about this project.
How good is the evidence that salt is actually bad?
Did anyone check if the people in this country eat too much rather than too little salt?