This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
Yeah, but surprisingly poorly. 2024-era LLMs can be prompted to play chess at amateur to skilled amateur levels, but to get to the superhuman levels exhibited by doing move evaluations with a chess-specific neural net, you need to train it using self-play too, and to get to the greatly-superhuman levels exhibited by the state-of-the-art chess neural networks of several years ago, you need to also combine the neural nets with a framework like Monte Carlo Tree Search. Just pushing human data into a neural network only gets you a third of the way there.
I'd guess that the "just pushing human data into a neural network only gets you a third of the way there" rule of thumb applies to a lot more than just chess, but it's a lot harder to "self-play" with reality than it is with chess, so we can't just make up the difference with more core-hours this time. Using "reasoning" models has helped, a little like how tree search helps in chess, by allowing models to try out multiple ideas with more than just one token's worth of thinking before backtracking and settling on their answer, but with a chess or go tree search there's still a ground truth model keeping things from ever going entirely off the rails, and reasoning models don't have that. I'm not sure what the AGI equivalent of self-play might be, and without that they're still mostly interpolating within rather than extrapolating outside the limits of their input data. Automation of mathematical proofs is perhaps the most "real-world" area of thought for which we can formalize (using a theorem language+verifier like Lean as the ground truth) a kind of self-play, but even if we could get LLMs to the point where they can come up with and prove Fermat's Last Theorem on their own, how much of the logic and creativity required for that manages to transfer to other domains?
MuZero would like a word.
At one point during training, the training environment was needed to keep MuZero from going off the rails and making illegal moves. Once it learns the rules of the game well enough, though, the policy network becomes sufficiently unlikely to output illegal moves that I expect it would continue to improve indefinitely through self play without sampling any illegal moves.
I do wonder if anyone has tried that experiment. It seems like it gets at the core of one of the cruxes people have about recursive self improvement.
More options
Context Copy link
More options
Context Copy link