@sodiummuffin's banner p

sodiummuffin


				

				

				
2 followers   follows 0 users  
joined 2022 September 05 03:26:09 UTC

				

User ID: 420

sodiummuffin


				
				
				

				
2 followers   follows 0 users   joined 2022 September 05 03:26:09 UTC

					

No bio...


					

User ID: 420

It's easy to forget today, in an era where so much of religion has retreated to views like "separate magisteria" and "well I guess God acted through evolution", but for millennia the existence of life was the best evidence for God's existence. Not "we can imagine a version of God that doesn't directly contradict the evidence" but "God is an intuitive explanation for this obvious phenomenon that we don't otherwise have an explanation for". Why do people and nature both clearly have so much purpose - eyes to see, teeth to eat, hands to hold, minds to think? When you drop some pebbles on the ground they don't assemble themselves into a sculpture and then start walking around, how did it happen with life?

An obvious answer is because someone chose to make them that way. That wasn't the right conclusion, even then: guessing a designer with a mind like we have was too much of a leap and equating that with [your culture's specific religious beliefs] was a much bigger leap. And while nature bears countless marks of purpose it also has a lot of elements that don't really make a lot of sense as a deliberate creation, especially not a deliberate creation by a single being with a unified purpose (for the monotheists), and especially not an benevolent one (this is presumably the inconsistency that the Garden of Eden myth was created to help explain). The right answer was "I don't know." But when one view has a specific intuitive explanation and the other has "maybe it was designed that way by someone unknown, maybe all life just appeared spontaneously the way maggots spontaneously appear in meat (another thing I can't explain), maybe there was some other unthinking process that I don't understand" it is understandable that people went with the first explanation.

The theory of evolution changed that. Suddenly there was a real explanation other than God, not just "I don't know but the burden of proof says...". Now, you might argue that as a practical matter the cat is out of the bag and writing arguments against evolution isn't going to convert millions of people to Christianity, but people don't think like that. Scripture provides non-evolutionary explanations for life and for centuries those were considered straightforwardly factual, not something to be excused away with "well the 'days' in which God created those things are a metaphor for the millions of years during which God was invisibly guiding natural processes such as evolution". It makes sense that some Christians would continue to argue on behalf of the explanations that were a core part of Christianity for so long, and with evolution as the dominant alternative that means trying to find arguments against evolution.

This is a pretty AI-savvy forum. Arguments like "Pangram is itself AI so it's bad/unreliable/hallucinating." or "This other AI detector didn't work and they're all the same, right?" may work on Reddit, but I don't think very many people here find them convincing. Pangram has been under significant scrutiny lately, aside from their own studies people have been putting quite a lot of effort trying to find false-positives without success outside of very short passages, low detection scores, and a poem where the same 6 words are repeated four thousand times. I know it's embarrassing to admit to undisclosed AI use, and dramatically more embarrassing after you've already denied it, but what you're doing now is, like the initial denial, just making it worse. Knowledgeable people reading your posts know that you're lying or, at best, using something like Grammarly or Copilot In Word that you don't realize is AI.

If you really are the first guy whose writing style creates Pangram false-positives, find a pre-2020 false-positive and collect your $4100. I predict that won't happen and that, like every other case of "Pangram thinks my writing style is AI", your writing only started detecting as AI after high-quality text models became available.

Pangram says 100% AI and it's obvious AI writing style, who are you trying to fool?

If your next argument is that Pangram is prone to false positives and you just happen to write like an AI to both human and Pangram's perceptions, note that there is currently a bounty of $4100 ($500 from this guy plus others offering their own funds) for a lengthy false-positive written before AI was available. Maybe some of your pre-AI writing is worth money?

https://x.com/Nition/status/2115166531462259163

I'm offering a USD$500 reward to the first person who can present pre-2020 human-written text that the latest version of @pangram flags as AI.

The specific requirements to claim the reward are:

  1. At least 750 consecutive words.
  2. Pangram flags it as 75% or more AI generated.
  3. You provide proof that the text was first written in 2019 or earlier and is unmodified.
  4. The text was ostensibly human written. This requirement only exists to avoid trick answers like output from GPT-2, machine translation, markov chains, the same word 500 times etc.

Don't use AI to write your posts, it's annoying to read and I don't want to dredge whatever point you were trying to make out of the morass of AI filler.

There was a guy on the Slatestarcodex subreddit that, based on the idea that some activities were more efficient at combining multiple benefits, tried to figure out what activities were most efficient. His candidate for the most optimal activity was playing sports outside in the sun with your friends.

The foremost benefit of sports is not "self-confidence", and it certainly isn't providing an advantage in zero-sum college admissions (which could likely be revised directly if you had the power to abolish women's sports). The foremost benefit is fitness, and secondary benefits include recreation, both spending time with people you know and meeting new people, etc. Competition provides a framework that encourages all of those benefits.

So why are middle and high school sports this one, weird exception where everyone wants segregation?

Middle and high-school sports? What's with the age-segregation? Why shouldn't middle-schoolers try to compete in professional adult sports leagues along with everyone else? The answer, of course, is that this would amount to them not having the benefits of organized sports at all. Abolishing women's sports is the same, but for half as many people.

You might ask, but wouldn’t this stop girls from playing sports? No, I don’t think so. They’ll still be on the team, they’ll still practice and compete. It just might stop them from being starters, because they’re probably not good enough for that. But that’s the same problem that most boys have, too. It’s not some fundamental human right to have a starting spot on the varsity basketball team.

This seems to contradict your other post:

It's disengenious to talk about how the boys' league is "technically open to all" when we all know darn well that will never happen.

In any case, it's wrong. Open leagues are both unrealistic and sometimes dangerous for women to compete in, if that wasn't the case far more female athletes would participate in the current open leagues for the bragging rights, fame, and superior competition. Indeed, it's already fairly common for female professional athletes to compete against high-school boys teams as practice, because that's the sort of gap needed to make it reasonably even.

The girl who was the star of her girls’ field hockey team gets a full-ride scholarship to Harvard, then gets recruited into a hedge fund, where she is quickly promoted to make diversity quotas. So she becomes an extremely confident, ego-tripping, “girl-boss,” with immense power to make bad decisions that damage the world, or at least her company. She makes life hell for her subordinates, and loses all attraction to any normal men. She advocates strongly for feminist programs that benefit women like herself: more money for girls sports, no trans girls allowed in girls’ sports, and more female executives.

This seems like something you should check before assuming, because I'm pretty sure it's all completely wrong. For starters, this study found no correlation between women's sports participation and attitudes towards sexism except in highly inegalitarian countries such as Iran. And of course opposition to MTFs in women's sports is more popular among right-wingers and thus would negatively correlate with support for diversity quotas. Also, I think if you took away their hobby in the name of anti-feminism that would make feminism more popular among women, not less. (Compare to SJWs censoring videogames in the name of feminism - how successful has this been at making more men feminist?)

Scott post about the paper from when it came out in 2018:

SSC Journal Club: Dissolving The Fermi Paradox

Asian-American women are not OK, at least if you go by what they write about.

Well, what they write about in "notable" books, which often includes going through the funnels of getting published and winning awards in an environment where [minority] writers talking about [minority] issues is highly prized. It's entirely plausible that if you went by either raw numbers of asian-american writers or by aggregate sales they're mainly writing self-published romantasy or the like. Part of the reason for the sameyness and triviality is that asian-americans don't have much in the way of potential racial grievances, especially if affirmative-action is off-limits.

The shared element is that both "only had an iced coffee to eat today" and "worried about Jew's apocalyptic cult that consorts with prostitutes and preaches universalist morality" are implicitly being presented as remarkable by the fictional interlocutor but are actually mundane and uninteresting because they happen all the time (where all the time in the second case means at least twice). It does add a major new element with the comparison though, which makes that less clear.

It's largely the same wording with the subject swapped out, same as these:

https://knowyourmeme.com/memes/should-we-tell-everyone-should-we-throw-a-party-should-we-invite-bella-hadid

Like many of those examples it isn't completely identical, it doesn't include the "should we tell everyone", doesn't elongate the "Oh", and adds "you" to "throw a party", but it's obviously still based on the template.

Oh you [statement]? Should we throw a party? Should we invite [related figure]?

aggressive in tone

Because it's a copypasta template-meme. In the original the tone contrasts with the frivolity of the subject-matter, but swapping that out for a more serious-sounding subject makes that part not work as well. Both are expressing exaggerated contempt towards someone saying something unremarkable as if it's remarkable ("I only had an iced coffee to eat today" and "A Jew's weird apocalyptic cult is overly universalist and hangs out with the outcasts of society") but the second makes the exaggeration less obvious.

Huh, looking him up he is the current president of MIRI so he does lead a rationalist institution even though he isn't particularly prominent among the online LessWrong community. He's just her boyfriend though, and looking it up they were already dating when he was just executive director at MIRI. I did say "leaders" plural, I was thinking in terms of people like Eliezer or Scott either hiring her services or using their positions as a leaders of the "rationalist cult" to get her to sleep with them, not her dating a guy who isn't particularly well-known but who rose to president of MIRI after they were already dating. I should have just said "Are you under the impression Eliezer is sleeping with Aella?" since I wasn't actually familiar enough with Aella to know who she was dating.

Some have pointed out that this is Alexander doing the meme

No it isn't. Doing the meme would at minimum require that that the tweet be an argument directed at Christians. It isn't an argument but a copypasta-template joke (with the humor partly based on the incongruity of the comparison) and the target audience for the joke isn't Christians (note that only 23% of ACX survey respondants are any kind of theist). Like his Antichrist post, which (perhaps because it wasn't posted on Twitter and was longer and thus harder to misinterpret) did not attract critiques about how it was an ineffective way to convince Christians that Marc Andreessen is the Antichrist. Or Unsong, the entire webnovel he wrote based in large part on similar jokes.

If I was to attempt to extract an implicit argument from the joke, it still wouldn't be one directed at or referencing contemporary Christians, but referencing the early Christians, as discussed in his post The Early Christian Strategy. The potential point would not be that EAs and internet-rationalists are followers of Christian beliefs, but that they have some similarities as weird rising altruistic movements that have been more successful than you might expect.

Scott's argument is a contemptable language game.

It's definitely meant to be a language game of the sort Scott frequently enjoys.

Jesus did indeed associate with prostitutes; he did not partake of their services, nor did he employ them on their backs for horizontal forms of evangelism-by-the-hour.

Uh, are you under the impression that there are internet-rationalist leaders sleeping with Aella? As far as I know Aella is just a normal prostitute/internet-celebrity who happens to also be interested in LessWrong stuff and do internet surveys.

I do not accept that the Yudkowskiites or the "Effective Altruism" movement they have parasitized are offering a "message of compassion"

To be clear the context for that part of joke is the Twitter arguments about EA which critique it specifically on the grounds of directing altruism at far people in other countries, people who aren't born yet, or non-human animals, as contrasted with normal healthy morality which is directed at family/community/nation. It's not trying to assert that EA is compassionate towards everyone (though Scott believes it is) but jokingly responding to people criticizing it on that basis.

A summary of her original written statement to police was just released today. Her statement to police does not allege that she was unconscious or was forced to have sex or take ketamine. She describes her refusals as being respected:

“[Male Number 1] would eventually wake up” and “said he wanted to take part and asked about making it a foursome, to which [Jane Doe] said ‘Hell No.” Jane Doe rejected this request “because [she] knew that a foursome meant that [she] would have to take part in anal sex.” Jane Doe then remembered Male Number 3 leaving the room.

And them stopping as soon as she said that she no longer wanted to have sex:

“The threesome continued until at some point, around 5:45am, [Jane Doe] said [she] was tired and needed to turn in. At that point, both guys stopped the threesome with [her] as far as [she] can remember.”

This is how she describes the ketamine use:

Male Number 2 “asked [Jane Doe] if she wanted to do more Ketamine.” At that, Jane Doe remembered “that [she] had taken Ketamine at some point on the couch with [Male Number 1 and Male Number 2] during this night” but she could not remember “where or when that took place.”

Jane Doe thought that she “had already done Ketamine once so why not do it a second time.” “At that point, [Male Number 2] poured a line of Ketamine on his erect penis, implying for [her] to snort the line. [Jane Doe] snorted the line of Ketamine, which was a heavy amount of Ketamine.”

I'm guessing that her going to the police was under the idea that intoxicated sex is rape even if she verbally agrees. However that does not actually constitute rape under New York law, which is why no charges were filed. Note that, for those who classify drunken sex as rape regardless of the law, reading the statement summary makes it seem like some of the other participants were probably drunker than her:

The three engaged in a “threesome” involving oral and vaginal sex in a room of the fraternity. At the start of the threesome, Jane Doe described her intoxication level as “5 or 6 out of 10” and that she was “pretty ok with having this threesome.” The sexual conduct ended when Jane Doe “noticed that [Male Number 1] appeared really sleepy and was no longer fully erect.” She told Male Number 2 “[they] needed to stop having sex in order to make sure [Male Number 1] was ok. [Male Number 2] then stopped having sex with [Jane Doe].”

Male Number 1 said “he was going to take a nap and refresh himself” and “would eventually stumble his way from the couch to the bed.” While Male Number 1 slept, “[Male Number 2] began to dry heave as if he was going to vomit. Jane Doe “began to take care of [Male Number 2] by patting his back while also waking [Male Number 1] every few minutes to make sure he was okay.”

In other news, per TMZ one of the accused was just fired from his job at Wells Fargo.

This is an example of how results based on "controlling for confounders" are inherently dubious for complicated fields like social science. Fundamentally the researchers are coming up with a list of potential confounders (which must be legible across many individual cases, available to the researchers, etc.) and assuming that anything left after they're controlled for must be whatever signal they're looking for. I recommend this SSC post as a classic example of this, in which a study on the impact of bullying controls for bullies targeting unpopular disadvantaged kids using the Bristol Social Adjustment Guide. In this case this is what they controlled for:

The model presented in Table 6 includes only case characteristics that would have been known by judges and jurors: the number of prior convictions; the number of victims (included here as a binary variable for one victim/multiple victims); the nature of the defendant’s plea; the number of applied aggravators (as determined by the judge or jury); the number of mitigating circumstances identified; the number of defenses offered; and whether the victim was held hostage. (We also tested the significance of a concomitant sex crime and whether the victim was a law enforcement officer; neither of these factors was found to be a significant predictor of sentencing outcomes and are not included in the model shown below).

This misses nearly all the details of the case. They count the raw number of official "applied aggravators" but not what those aggravators are. They count the number of prior convictions but not their severity, the number of mitigating circumstances and number of defenses but not what those are or how convincingly they are argued. They check whether the victim was held hostage for some reason but not other similar specifics (chosen controls tend to be arbitrary and vary widely across studies on the same subject). The people making the decision have all the details that a set of controls can't possibly encompass, and we know the controlled factors differ by race, yet any systematic difference in their decisions is assumed to be a product of racial discrimination rather than everything they didn't control for.

Imagine a study trying to determine if the date of birth on a birth certificate is a biased indicator of age. Group 1 is stereotyped as being older than Group 2, does this influence the date written down when they're born? Naturally the study controls for age, with a multiple-choice question to choose an age range ("[B]: 20-29 years old"). They find that, while Group 1 does pick older age ranges, even after you control for age [B] respondents in Group 1 have an earlier date of birth than [B]s in Group 2. In fact, a later more sophisticated study outright asks for age, and finds that 25 year olds in Group 1 have an earlier birth date than 25 year olds in Group 2. Needless to say, this is a perfectly expected result even without any birth-certificate bias.

Interestingly, after controlling for race of defendant, the number of mitigating circumstances is no longer a significant predictor of decisions to impose death.

In other words "race" and "number of mitigating circumstances" correlate strongly enough that if they didn't put "mitigating circumstances" in their model any difference in sentencing from mitigating circumstances would be assumed to be due to racism. Naturally they don't consider whether the same thing might be happening with factors they didn't control for.

Second, the regression results indicate that case characteristics explain only a small proportion of the variation in the case outcomes analyzed here. Specifically, case characteristics alone explain only 9% of the variation in prosecutorial decisions regarding whether to seek death and 20% of the variation in juries’ sentencing decisions.

Unexplained variation documented in the results presented here suggest that other extra-legal and social factors –not captured by our statistical models – are playing an important role in death penalty case dynamics.

Interestingly their conclusion calls the variation not explained by their model "extra-legal and social factors". I don't think it's really "extra-legal factors" when the jury makes decisions using methods more complicated than "count up the number of applied aggravators".

All three of OpenAI, Anthropic, and xAI were founded by people who were already expressing concerns about AI wiping out humanity before modern AI existed. OpenAI started in 2015 as a nonprofit with the aim of making sure AI benefits rather than harms humanity. Anthropic was founded in 2021 in the GPT-3 era, before ChatGPT, by OpenAI employees who thought OpenAI wasn't taking AI safety seriously enough. They have been predicting AI risk continuously for years and they have had intimate experience with lesser predictions about AI advancement coming true. It is crazy to me how reluctant people are to believe they are sincere about worrying that careless AI advancement will kill them and everyone else, rather than it being a plot to make money when they're already rich.

Come on dude. EA organizations make and publish analyses of effectiveness all the time, it's their whole thing. Here is an EA blog post about bednet misuse I found in a minute of searching. As it mentions, Givewell commissioned Rethink Priorities to research use of bednets in fishing:

Based on the available data, Rethink Priorities estimates that only 1.2% of bednets globally are repurposed for fishing.

Imperfect use is already included in cost-effectiveness:

For example, GiveWell estimates that only 60% of recipients will use their nets consistently and correctly. After accounting for this, they still find bednets to be extremely cost effective, with every thousand nets saving a life in expectation. That 400 will not be used correctly— and that 12 might be used for fishing— is already baked into the calculation.

And of course there is a ton of research, including randomized control trials, showing distributing bednets is effective:

The typical study design involves randomly assigning groups of people to receive (or not receive) nets. Researchers then wait to see whether mortality and malaria rates differ between the two groups. There is never an assumption that everyone who receives a net will use it. And yet, such studies consistently find that nets reduce child mortality by around 17 percent.

On his death, there were many in the Right who believed he could serve as the martyr that would finally force the Right to "get serious". This meant very different things to different people from retaliating for cancel culture, to aggressively implementing Trump's immigration agenda, to turning the US into an outright dictatorship and instituting the Day of the Rope. In the end, none of this happened since his death got bogged down in endless conspiracies.

No, none of that happened because...why would it? Politicians who want conservative policy are generally already doing their best to implement it, though sometimes their best isn't very good. It alienated some left-wingers and moderates to see celebration of the assassination be relatively mainstream in left-wing spaces, but it's not something that was going to cause a huge shift in either public opinion or politician's opinions on any specific issue. Being a left-wing assassin is already illegal. There's been more of a crackdown on antifa groups - probably the Autistici/Inventati Collective wouldn't have gotten designated as a terrorist group (and officially dissolved 4 days ago) if it wasn't for Kirk's death - but most antifa activity is already either clearly illegal or clearly legal and also antifa doesn't actually matter that much.

Regardless of what improvements can be made to daycare regulations, "childcare subsidies but only if you pay for someone else to do it" is distortionary and produces economic inefficiency. Even if you hypothetically fixed regulations so hard that paying for daycare was always cheaper than even the worst-paying jobs, it would still be paying people trading off between "more money so we can save up for a cruise" and "getting to spend time with my kids instead of in an Amazon warehouse" to take the non-utility-maximizing option.

Making the childcare subsidies less distortionary is not "more government" in any meaningful sense, especially since it's coming out of the same fund so it is literally the same number of tax dollars being distributed. And of course until such hypothetical improvements are made we get the current situation where daycare for a couple children often costs more than the entire take-home pay for one of the parents, but the government will pay people to use it.

And in a fair fight it's very obvious that daycares win economically because one person taking care of ten children frees up nine workers.

I think you are underestimating the overhead inherent in things like property costs, having a formal organization where people can sign up their kids, and compensating for the lower trustworthiness of multiple for-pay non-parent carers. Regarding the last I'm reminded of the statistic that non-relatives in the home are massively over-represented in abuse, while obviously some parents are abusive or neglectful it's less common. Not to mention the simple fact that more carers means a higher chance that at least one is bad. So if you're having strangers take care of your children there's an incentive to pursue overhead-increasing safety measures and expensive signaling like "all carers have a college degree in Early Childhood Education" or "daycare seems very well-maintained and high-class" to reduce risk, regardless of regulatory requirements. In any case, whether it can be made dramatically cheaper or not, subsidizing it over unpaid childcare is fundamentally paying people on the margin to choose it even if it would otherwise be the worse choice.

We have a much better way of measuring whether it cancels out than the number of children per carer: the number of dollars. If some of the dollars instead go to "the construction workers who built the daycare building" and "the daycare administrators who handle the legal issues and overhead of taking care of other people's children" that doesn't represent an increase in efficiency. If anything having a lower carer to child ratio will on average correspond to lower quality (not to mention differences like how on average carers care less than parents do), so if daycare costs more than Alice's job pays she is paying more for lower quality. Rather than deciding on Alice's behalf that daycare must be more efficient because it has more children, we can let her compare her wages to the cost of daycare (not to mention other factors like how much she likes her job) and decide for herself. Except that, by subsidizing paid childcare, we distort this comparison and end up paying her to take the economically less efficient option. Equally subsidizing childcare whether she does it herself or pays a daycare to do it removes this distortion, letting her once again decide for herself based on her individual situation which is the best option.

So if Alice and Susie are paid to take care of their own children they are welfare leeches, but not if Susie is paid to take care of Alice's child and Alice is paid to take care of Susie's child?

If Alice is paid less than the cost of daycare for working, would do a better job than the daycare, and would prefer spending time with her own children to spending time at work, then working that job is not economically efficient. Subsidizing paid childcare is thus an obvious example of creating economic inefficiency, and the stay-at-home parents who get screwed by their exclusion have understandably been complaining. (Subsidizing childcare in general isn't necessarily inefficient because the average person creates net-positive externalities which parents aren't paid for because it's illegal and impractical to sell your children into slavery or take out loans against their future earnings. Thus the job of creating people suffers from underinvestment.)

Now, it is possible that subsidizing daycare but not stay-at-home parents served a beneficial purpose despite the obvious economic inefficiency. The average first-world person has net-positive externalities but lots of people do not. I assume that is what some of the backlash against "welfare-queen"-enabling policies was, not just enabling people to be net-negative to societal productivity but paying them to create more people who will likely also be net-negative in a fashion that is obviously unsustainable because somebody has to actually pay taxes. Proposals to increase the fertility rate often take the form of reducing taxes rather than direct payments for this reason. However the new proposal would require that the spouse be working 35+ hours a week, which might be sufficient to make it net-positive despite it cutting out at higher income levels.

Fundamentally, nerds are more likely to hold non-mainstream political views because they are more likely to have picked up their political ideas from the internet, from other nerdy sources like political books, or from nerd peers who got their views from nerdy sources. Non-nerds are more likely to have gotten their views from sources like school, TV news, or non-nerd peers. Because non-nerd sources were more centralized they were more likely to be controlled by the political memeplexes that were already the most powerful, like mainline Democrats/Republicans, and not include niche or taboo views. The internet, on the other hand, favors views that are well-optimized for grassroots virality (with a secondary advantage for being good at taking control of website moderation, though this can also spark backlash). There may be psychological reasons why people favor the exact views they do, but first and foremost any analysis should keep in mind the viral nature of ideas.

Of course, this dynamic is changing as more of the population is heavily online, though there is inertia both on the individual level (people tend to stick with the views they are already committed to) and the group level (views that are already dominant in a social group have an advantage). Increasingly the nerd/non-nerd division is not between online and offline but between more and less nerdy websites (ordered something like political-blogs>Tumblr>4chan>Reddit>Twitter>Tiktok, though some of the biggest political blogs might have a less nerdy audience). As a result, niche and taboo views are increasingly popular among non-nerds, though the kinds of views which gain popularity on Tiktok are different from those that gain popularity from long-form political blogs. Furthermore, the internet provided a space for the development and refinement of once-niche views which in turn managed to seize control of (or at least influence) more mainstream institutions like news outlets and political parties. So even people who still get their views from the news and mainstream institutions tend to just be downstream of whatever internet memeplexes have managed to take over those institutions. This includes the linked article, which is essentially an extension of an internet flame-war where the ideology which now runs NPR is using it to take shots at those tech CEOs who have been converted by enemy ideologies.

Most of them were involved in fundraising that, if done competently, involved a due diligence component. So either they were incompetent, or chose to interact with the publicly convicted sex offender anyway.

I think it's the latter. The AI research community is deeply entwined with both rationalism and Effective Altruism, and the latter is a rehash of utilitarianism with a peculiar utility function. Artificial super intelligence, if correctly formed, presents infinite utility. If that is true, then literally anything can be justified if it moves the future toward that end.

If Dario's wife was asking him for funding because 'AI justifies anything', she would presumably have asked for AI funding rather than trying to fundraise for a completely different company. Meanwhile Eliezer Yudkowsky is connected to Epstein because he accepted a phone call during a MIRI fundraiser...and then they turned down his $300,000 donation offer after doing due diligence. Here is a comment from him describing the interaction:

Epstein asked to call during a fundraiser. My notes say that I tried to explain AI alignment principles and difficulty to him (presumably in the same way I always would) and that he did not seem to be getting it very much. Others at MIRI say (I do not remember myself / have not myself checked the records) that Epstein then offered MIRI $300K; which made it worth MIRI's while to figure out whether Epstein was an actual bad guy versus random witchhunted guy, and ask if there was a reasonable path to accepting his donations causing harm; and the upshot was that MIRI decided not to take donations from him. I think/recall that it did not seem worthwhile to do a whole diligence thing about this Epstein guy before we knew whether he was offering significant funding in the first place, and then he did, and then MIRI people looked further, and then (I am told) MIRI turned him down.

Epstein threw money at quite a lot of scientists and I expect a majority of them did not have a clue. It's not standard practice among nonprofits to run diligence on donors, and in fact I don't think it should be. Diligence is costly in executive attention, it is relatively rare that a major donor is using your acceptance of donations to get social cover for an island-based extortion operation, and this kind of scrutiny is more efficiently centralized by having professional law enforcement do it than by distributing it across thousands of nonprofits.

In 2009, MIRI (then SIAI) was a fiscal sponsor for an open-source project (that is, we extended our nonprofit status to the project, so they could accept donations on a tax-exempt basis, having determined ourselves that their purpose was a charitable one related to our mission) and they got $50K from Epstein. Nobody at SIAI noticed the name, and since it wasn't a donation aimed at SIAI itself, we did not run major-donor relations about it.

This reply has not been approved by MIRI / carefully fact-checked, it is just off the top of my own head.

So the only example in your list of people of Epstein funding AI is the money he gave to MIT Media Lab and MIT AI Lab, which is connected to academic AI research but not to internet-rationalism or effective altruism. Also I suspect their policies towards accepting donations has much more to do with MIT-wide policies than anything to do with AI specifically. The only internet-rationalist organization is also the one that got a donation offer and turned it down. Rather than anything to do with rationalism, EA, or utilitarianism, I think this just reflects AI being one of the fields that Epstein thought was cool.

Most people die for normal reasons, like terminal illness. If you convert a significant fraction of those deaths into MAID deaths, it is unsurprising if MAID deaths end up outnumbering abnormal causes of death like getting shot. This is pretty irrelevant to the question of how many MAID deaths are bad. (People disagree on exactly which MAID deaths qualify as bad, but "terminal cancer patient in pain chooses to die a couple weeks early" is pretty popular and "person who could have lived a long happy life successfully pursues MAID based on some bullshit condition because it gives institutional approval for suicidal tendencies" is not.) On one hand the raw number of MAID deaths doesn't indicate abuse of the system, on the other hand a high percentage of reasonable MAID deaths doesn't tell us whether or not there is a significant amount of abuse, since you would expect the normal deaths to be the majority regardless.

It is the drowning child thought experiment brought to its logical conclusion.

More like a quarter of the way towards its conclusion. Low-productivity migrants are very expensive: this blog estimates the net lifetime tax contributions of Somalian migrants in Denmark to be around negative $1.4 million USD. Meanwhile, Givewell estimates that preventing a death from malaria costs $3000 to $8000. For the cost of improving the life of a single Somalian by allowing him to migrate, you can outright save the lives of hundreds of other africans who remain in Africa. (Before you object that saving the lives of poor third-worlders just results in more third-worlders who need saving, note that it apparently only costs $36 to a contraception charity per third-world birth prevented.) Poor third-world countries have a lot of low-hanging fruit where small amounts of money go a long way, so it shouldn't be too surprising that helping third-worlders in their own countries is vastly more efficient than letting them migrate.

The drowning child thought experiment says it shouldn't matter if the child is in front of you or in another country, which is quite unlike the instinctual morality that says cheap foreign aid programs saving millions of lives is a minor side-issue that only becomes morally urgent when one of them manages to get closer and throws himself into the water. Sometimes literally, with those ocean-rescue NGOs that transport migrants to Europe if they go out onto international waters and in some cases deliberately sink their own boats. Old 2016 article, I haven't kept up with how the situation has changed since then:

Time: Don't Blame Rescue-at-Sea Organizations for Migrants Coming to Europe

When 22-year-old Mohammad left Libya’s coastline in a decrepit wooden boat headed for Italy in August, he had no illusions about where he would end up.

Unlike many smugglers in Libya who lure clients in with false promises of reaching Europe within hours, his agent was brutally frank. It would take about three hours to reach international waters, he explained. Once there the boat would drift until spotted by one of the humanitarian rescue vessels plying the central Mediterranean, at which point he and his fellow passengers would be taken aboard and transferred to Italy, where they could apply for asylum.

This is the opposite of trying to maximize the most good for the greatest number in a location-insensitive way. It's like a thought-experiment someone invented to show how certain implicit moral premises are incoherent and produces counterproductive results, except that people are doing it for real. As for what actually qualifies as "the drowning child thought experiment brought to its logical conclusion", I suspect it's some longtermist thing where you realize that the number of people in the future is potentially large and helping them by keeping society from collapsing and advancing technology is dramatically more efficient than anything else. That has a lot more uncertainty though, so providing Africa with bednets and contraception is still fairly reasonable, unlike helping them by allowing migration.

If you fund sub-Saharan Africa with USAID you get more Sub-Saharan Africans.

Is this true? Typically lower mortality and higher quality of life are associated with lower birth rates. There are also various programs to lower birth rates more directly, like providing access to contraception and educating women, which could plausibly outweigh any increase in births due to saving lives.

A quick search finds a study saying "USAID was one of the largest providers of family planning services worldwide with almost $600 million for contraceptive services in poor countries." and a New York Times article claiming that $8.1 million in contraception was ruined after being stranded in Belgium due to the cuts. I do not know how much of these programs ultimately ended up being cut vs. being moved around.

This Givewell analysis estimates that each year of contraception prevents 0.3 unwanted pregnancies, of which 45% result in birth. This comment from someone at MSI Reproductive Choices says their average global cost per year of protection is $4.88, though increased marginal access could in some cases be $20 or more. So preventing a birth apparently costs around $36 for a private charity. Obviously USAID contraception could be less efficient than EA-adjacent charities, or more efficient due to increased economies of scale, but just multiplying it out implies USAID directly prevented the birth of around 16 million third-worlders per year, and that the one stranded Belgium shipment meant for Africa resulted in an additional 200,000 africans being born (or more because $8.1 million was the cost of the physical contraception, while $4.88 includes last-mile delivery). Realistically some of those averted pregnancies will probably be delayed rather than prevented entirely, since people might get access to contraception and use it based on how many children they already have, but it gives a rough idea. The Lancet study claims cuts would result in additional 14 million deaths but that's over 5 years, over the same period USAID contraception programs are (by these rough estimates) preventing 80 million from being born. Of course the Lancet's death estimates are based on assuming cuts to programs like PEPFAR that were actually moved to a different department, and I can't easily find whether the same happened with the contraception programs, so I don't know how it actually balances out.

I think this demonstrates the nature of low-hanging fruit/"neglectedness" in international aid and the importance of even very rough numbers over going by vibes. Because places like Africa are very poor, there is a lot of low-hanging fruit. Because more people are interested in saving lives than preventing births, there is even more low-hanging fruit in the latter. It's $3,000 to $8,000 to buy enough anti-malarial bednets to save a single african life, but only ~$36 to improve the average well-being of humanity by preventing someone from being born in Africa to begin with. Even though contraception was only 1.3% of USAID's $47 billion pre-cut budget, it seemingly had more of an impact on third-world population numbers than the rest combined. Spending fairly trivial amounts of money on increased contraception funding for the third-world could have a much bigger impact, save money long-term, and be much less controversial than the USAID cuts. Of course left-wingers are not immune to this sort of innumeracy either - consider how estimates of net lifetime costs to the taxpayer for Somalian migrants is over a million dollars per migrant, because doing things in the first world is expensive. Instead of improving the life of a single poor african by importing him to the first world you could save the lives of hundreds of africans by spending the money on insecticidal nets. And then because birth-control is even more neglected, for the cost of saving a single life from malaria you could potentially prevent over a hundred from being born.