Adrien Ecoffet

thinking claude is conscious: a viewpoint i disagree with and think is a little silly tbh thinking claude fundamentally can't be conscious because it's a computer: the absolute height of comical midwittery

Chess is 30 years ahead of every other profession in dealing with AI. The best case study we have for what's coming. 4 lessons: 1. Human-AI collaboration had a 15-year shelf life in chess. "Human in the loop" is a phase.

@kamathematic “update your priors” = change your ecclesiastical superior “orthogonal” = relating to a polygon with ortho many sides “isomorphic” = currently transforming into the international standards organization “overfit” = too healthy “steelman” = the tin man's cousin

I won't really be impressed with AI until a dog uses ChatGPT to cure a human's cancer

Warning: Do not adopt any new code editors this month. Beware the IDEs of March.

im working on it okay

@ForHumanityPod Why does this person need to be a good golfer?

Serious question: Who is the best golfer in AI safety? Plan: -Get this person a Mar-A-Lago membership -Pay the $500k to golf w Trump -On hole 7 mention his only way to a Nobel Peace Prize is a an AI safety treaty w China -On hole 11 say if he did this even his harshest critics would have to love him. -Have media ready at hole 18, he walks off the course, says we need an AI safety treaty w China and kicks the Overton window open. This could really work.

Old enough to remember when "AGI" was a taboo word, and the idea that scaling transformers could get us there were downright offensive to many in academia

Two Diet Cokes Three Cokes Zero

1. Agree. 2. Disagree. AI improvement usually has power law scaling, not logarithmic. Even if the power law has substantially diminishing returns you still get an exponential with exponentially growing inputs. 3. Mostly disagree. Would take a long time to discuss in detail but I think most of these are actually solved or made irrelevant by more scaling. 4. Not sure enough what you have in mind to say. 5. Mostly Agree (high uncertainty). 6. Strongly agree. 7. Agree.

1. From a bayesian perspective, low p(doom) should be the default. The burden of proof is on those with a high p(doom). They have not made a case that convinces me. 2. AI recursive self-improvement models ignore so much hard stuff. In particular, that AI improvement shows logarithmic diminishing returns to inputs. Every linear step of AI improvement takes exponentially more inputs. 3. AI boosters, particularly of the "scaling is all you need" variety ignore the multiple AI challenges that need theoretical or algorithmic solutions: Incredibly low data efficiency for learning; Difficulty generalizing beyond the training set; Not knowing what they don't know; Continuous and real-time learning; Lack of introspection; etc... 4. Most (but not all) high p(doom) scenarios imply a level of volition around AI that I just don't see. As I said, not all. 5. I do think (despite OpenClaw and talk of giving Grok control over robot bodies) that we'll have some sensible limits in what real-world systems we give AIs control over. Or constraints around how they can direct those systesm. 6. The scale of harms AI can cause (accidentally or intentionally) probably have some sort of power law scaling. Most likely, for that reason and others, early safety mistakes will be small and not existential. And we'll learn from those and make the world safer and more secure. 7. We are not dumb as a species, and many smart people are working on how to increase AI safety. We'll make mistakes. Some bad things will happen. And we'll get smarter and fix the holes. That's what's happened with essentially every technology to date.

Who called it "Effective Altruism" and not "Crazy Rich Bayesians"?

I've developed the unique skill of being incredibly sharp in meetings that don't matter, and a half-zombified wreck when surrounded by people who actually do. Bodes well.

BREAKING: Andrew Yang says we should stop taxing workers, and tax AI instead. That can fund UBI. "We should try to stop taxing labor."

“a governance model that depends on every frontier ai company acting against economic incentives, by relying on voluntary commitments that can be changed unilaterally at any moment, is not the kind of safeguard a free society should rely on”

can’t believe you fell for the anti-california psyop just pay the taxes lil bro

I find it interesting to watch this video and look for analogies to the emergence of superintelligence in our civilization Many cells start acting in a coordinated way to become a heart. What would the arrival of ASI in our society look like from the perspective of the universe?