>Writing in Asterisk, a magazine related to effective altruism, Ozy Brennan criticized Gebru's and Torres's grouping of different philosophies as if they were a "monolithic" movement. Brennan argues Torres has misunderstood these different philosophies, and has taken philosophical thought experiments out of context.[21] Similarly, Oliver Habryka of LessWrong has criticized the concept, saying: "I've never in my life met a cosmist; apparently I'm great friends with them. Apparently, I'm like in cahoots [with them]."
The people that coined TESCREAL seem to not really be related to rationalists, and seem to have coined a term for "those vaguely related ideas from people that do some stuff we consider wrong and we consider bad". "Evil people from San Francisco" could work just as well I think.
And wait, shouldn't I beware arguments for utilitarianism rather than against? If that's what you meant yeah I agree, especially pushed to the extreme it leads you in very weird places.
I know nothing about the drama, but treating "utilitarianism" as if it is one thing or that a particular person or group's position is identical to utilitarianism seems ironic in the context. It is like claiming all pizza is bad because I went to dominoes and didn't like the experience.
I'm not sure what the parent meant by "beware arguments against utilitarianism" - there is nothing wrong with arguing for or against utilitarianism. It's a popular moral philosophy.
You should beware of bad utilitarian arguments though, which is where you often get the real "gotta break a few eggs to make an omelette" kind of arguments that justify all manners of atrocity in service of some narrow hypothetical future good.
Like when Marc Andreessen says we should consider anyone who would do something to slow down or regulate AI advancement murderers of future humans. Bad utilitarianism right there.
Proper utilitarians are concerned with the net difference between all positive and negative consequences of actions.
People often think they have mic-dropped utilitarianism by saying things like, "Oh, so if two people get a lot of joy by beating up a third person, that is ethical because it is overall net positive?"
A few things wrong with that. First is there is no net happiness formula which utilitarians are proposing. Peter Singer has said more than once that he weights suffering far, far higher than happiness.
Second is that every ethical system has screw cases which make the system look messed up. "Do unto others..." it terrible if you are talking about masochists.
Like any group of humans, there are power structures and edge cases that can lead to horrific outcomes. Giving the person that posted the warning the benefit of the doubt, I think what they are saying is that "Rationalist does not necessarily mean positive for humanity, nor even no harm for humanity". This holds for all religions and religion-like movements, of which Rationalism, in this sense, is one.
I think this holds for basically all movements in which case, I don't really understand the need to flag that website over any website.
Edit: though I don't want to diminish that this specific group is a cult with classical cult techniques like sleep deprivation and with the disastrous consequences often associated with cults.
> They claim to practice unihemispheric sleep (UHS), a form of sleep deprivation intended to "jailbreak" the mind, which they believe can enhance their commitment to their cause.[
> I think this holds for basically all movements in which case, I don't really understand the need to flag that website over any website.
Because it uses the language and thinking patterns of the Rationalists, it serves as a strong indoctrination tool. The site itself isn't bad but, as someone who flirted with those communities as a result of the site, I think the warning is deserved.
If you feel like the warning is deserved, and have personal experience, yeah then fair enough. I personally live in the "periphery" (read: not in Berkley or SF or New York) so I think I may see way less of the good and the bad.
It's a particular variety of "everyone else is wrong (and maybe a bit stupid)".
Like, sure, sometimes you get popular nonsense like recovered memories or accidental fires can't be as hot as intentional fires or shaken-baby syndrome or bite-mark analysis. But a lot of times, everyone isn't wrong and you've just overlooked something critical or misdefined the problem.
> Like, sure, sometimes you get popular nonsense like recovered memories or accidental fires can't be as hot as intentional fires or shaken-baby syndrome or bite-mark analysis. But a lot of times, everyone isn't wrong and you've just overlooked something critical or misdefined the problem.
The older I get, the more I find that everyone is wrong. It's fucking astounding how much stuff either was never actually checked, or is true only under very select circumstances with those caveats being widely ignored. For example at work right now we have been using a test for 40 years that was developed around the idea that our product absorbed air - chemical variation would lead to extreme differences in results and you can't retest an item for at least 24 hours because it will still be affected. Turns out that none of that was true, all the error we were getting was from temperature change, the items can be retested after 45 seconds. 40 years and no one took 30 minutes to verify this claim which costs us millions of dollars per year. And this is just the example from this past week. I've probably seen several hundred such cases of completely unjustified claims being treated as gospel truth.
I can't speak for the countless things I've never tested, but if nearly everything I do test is wrong, across numerous fields full of very intelligent people, it doesn't give me much confidence about everything else. We live in a world that values simplicity and confidence, not nuance and rigorous verification. I've gotten to the point where I don't trust anything without verification, not even my own past work.
> It's a particular variety of "everyone else is wrong (and maybe a bit stupid)".
Gestures at the current state of the world
Not that adopting rationalist modes of thinking will fix the problem, of course. Teach rationalist principles to an idiot and you will have a slightly more rational idiot, who will reason himself into absurdity. Teach them to a manipulative, amoral psychopath and you will have a more skillful manipulator.
Rationalist principles and methods provide superior tools for thinking through some complex problems, but they say nothing about foundational ethics (other than pointing out possible sources for the many different systems of ethical beliefs). And they cannot be wielded effectively by people who lack the ability to decouple, to think abstractly, or to create extended “chains” of thoughts and keep them in working memory.
One should be suspicious of anyone who claims that rationalism is a panacea, or alternatively that it is somehow a problem per se. It’s a neutral set of tools, a community who wants to improve those tools, and a small group hanging off the edge who have unrealistic and/or harmful views of how those tools should be applied. Unfortunately this third minority is presented by anti-rationalists as the core of rationalism. In reality, they are easily avoided unless you hold the same core values.
(I say this as a long time observer who appreciates their work but does not consider myself a part of the “rationalist” community.)
>a right-side political figure, who are basically ruling since 2000, (except from 2012-2017 where France had a social-democratic government and president)
I think it is important to try to find more rigorous things to test than the general sentiment of the people using the tools. If only because the more benchmarks we have the more we can improve models without regressions. METR is asking a really interesting question here, "are models improving at making one shot PRs?". The answer seems to be, yes, but slower than benchmarks suggest, if you look at the pass rate of different versions of Claude Sonnet. A reasonable answer is "you're not supposed to use them by making one shot PRs", but then ideally we would need to have some kind of standarized test for the ability of models to incorporate feedback and evolve PRs.
>To study how agent success on benchmark tasks relates to real-world usefulness, we had 4 active maintainers from 3 SWE-bench Verified repositories review 296 AI-generated pull requests (PRs). We had maintainers (hypothetically) accept or request changes for patches as well as provide the core reason they were requesting changes: core functionality failure, patch breaks other code or code quality issues.
I would also advise taking a look at the rejection reasons for the PRs. For example, Figure 5 shows two rejections for "code quality" because of (and I quote) "looks like a useless AI slop comment." This is something models still do, but that is also very easily fixable. I think in that case the issue is that the level of comment wanted hasn't been properly formalized in the repo and the model hasn't been able to deduce it from the context it had.
As for the article, I think mixing all models together doesn't make sense. For example, maybe a slope describe the increasing Claude Sonnet better than a step function.
That is true but also a bit unfair, they've also been oddly preoccupied with topics like trying to help the most people and frequently promote giving money to efficient charities to fight against malaria, vitamin A deficiencies and help vaccinate children in very poor countries.
This image comes from running the different versions of the benchmark games programs. Some of the difference between languages may actually be just algorithmic differences, and also those programs are in general not representative of most of the software that runs.
I have no tolerance for bystanders being killed in general. If the science experiments kill on average less bystanders I'm all for them, if they don't they should be stopped until made safer.
In this case the judgement is so extreme because the judge had no tolerance for Tesla lying in relation to the server logs' existence and what they contained (namely that is was indeed their autopilot that was in full control, had been in full control for almost half an hour, and was not worried at all/not issuing warnings, at the time of the crash)
What does that mean?
reply