https://www.theargumentmag.com/p/losing-control-of-ai-is-actually I'm going to go out on a limb here and say I think RSI is a bad idea. Yes, even though I've been vocally pro-AI in the past. In general I have a strong presumption that technological progress should be allowed to continue freely, and it takes something exceptional for me to admit an exception. I think this plan, for RSI in 2027, seems reckless. I will be relieved if it is prevented from going through.
https://www.anthropic.com/research/formalizing-fermats-last-theorem I knew about the project to formalize Fermat's Last Theorem back in 2024 and I did not expect it to complete this early. I definitely did not expect Claude to do the whole thing in 11 days. I haven't independently verified that this is real (and I can't, not knowing the relevant math or Lean), but if so it's extraordinary.
https://cims.nyu.edu/~tristanb/statement.pdf if I understand correctly, this is OpenAI playing dirty. OpenAI is trying to be "first" to a mathematical result, and is willing to more-or-less plagiarize mathematicians to get it. (using the chat transcripts of mathematicians working with an OpenAI subscription.)
it is also important to note that AI-generated math can be "slop" in many ways. It can be literally incorrect; it can be so messy and/or verbose as to be uninterpretable and uncheckable by humans, containing irrelevant tangents even if there is a valid proof somewhere in the mess; it can be meaningless (which is particularly important in topics such as the Navier-Stokes equations, which are derived from physics; if you choose a variation on the problem that has no physical meaning, and prove something about that, you are not necessarily doing something relevant to the main problem).
without a good faith collaboration with mathematicians, we will simply never know whether AI companies' announcements of AI-generated "mathematical results" are to be trusted. Plagiarizing mathematicians is not only unethical in itself but also destroys trust in a way that should call the whole enterprise of AI-company-led mathematics into question.
"I said that if OpenAI released its result in the way proposed I would go public with what happened. The reply was, “Why would you ruin your career?” I replied that I am an academic, and asked why he thought going public would ruin my career. The reply was, “If you don’t want me to be nice, then I don’t have to be nice.” WTF WTF WTF.
https://vaticidalprophet.substack.com/p/within-person-cognitive-variance this fits with my preconceptions, and brings a lot of data. cognitive variance between different tests is really wide, for both neurotypical and neurodivergent people. Yes, there are positive correlations between different cognitive skills (that's what "g" is) but "multiple intelligences" is almost more correct as a high-level intuition than "intelligence" as a single thing. most people are quite a bit better at some things than others.
I've said it many times and I still believe it; "people are really different from each other mentally -- not better, not worse, just different" is largely true and deeply counterintuitive, and even horoscope-y "personality type" systems are directionally good to the extent that they actually get people to believe it. when you have a disconnect with another person, you will tend to assume they are malicious, superior, or inferior -- it will be exceedingly hard to seriously entertain the hypothesis that they are a fellow human, roughly no better or worse than you, whose mind just works in an incredibly different way from yours. but it will often be true. More-or-less peaceful and more-or-less egalitarian coexistence, across very wide differences, is not a kindergarten fantasy, but a real necessity that you will wind up coming back to even after you accept the existence of evil, conflict, differences in ability, etc.