Running Lightcone Infrastructure, which runs LessWrong. You can reach me at habryka@lesswrong.com. I have signed no contracts or agreements whose existence I cannot mention.
Posts
Wiki Contributions
My model was that Claude Sonnet has tool access and sometimes does some tool-usage behind the scenes (which later gets revealed to the user), but that it wasn't having a whole CoT behind the scenes, but I might be wrong.
Oops, sorry for ruining your experiment :P
(Edit note: I imported the full text, based on a guess that you didn't post the full text because it was annoying to import, but of course feel free to revert)
Yeah, I also think you misunderstood that part of the paper (though it really is very ambiguously written). My best guess is they are saying that they are averaging their performance metrics over 5 forecasts.
2Thank you for this!
Edit note: I slightly cleaned up some formatting in the post which looked a bit janky on mobile (due to heavily nested bullet lists with images). Feel free to revert.
Double your impact! Every dollar you donate to the Center for AI Safety will be matched 1:1 up to $2 million. Donate here.
In all but the rarest of circumstances matching campaigns are pretty deceptive, and claiming that this would "double your impact" is almost certainly false. See all the standard critiques of matching campaigns.
Also, the donate link doesn't include anything about the matching campaign.
Yeah, I mean, I think it's fine and am glad about this work, I just definitely misunderstood (but it's all a very harmless misunderstanding).
I am not Daniel, but why would "constant growth" make any sense under Daniel's worldview? The whole point is that AI can achieve explosive growth, and right now energy consumption growth is determined by human growth, not AI growth, so it seems extremely unlikely for growth between now and then to be constant.
This all roughly matches my model.
Perhaps someone from the LW team who was around back then can better describe the details, but as I understand it, it was a design flaw in the “if someone does this bad thing, we have no good tools to catch them and/or prevent someone from doing it” sense.
Yeah, the core issue was that there was basically no protection against people just automatically creating thousands of accounts and using them to downvote whoever they disliked. In-general, bot issues are one of the top reasons for websites that accept user submissions either need to have a strict manual review phase, or be continuously updated with defenses.
What is the source of this observation?