@KelseyTuoc@LeroyTheElk They're almost never hundreds of pages. This one was only 6000 or so words (22 pages, but they have massive margins). There's also concurrence and a dissent which are another 1000 and 6000 respectively.
@bubbling_creek@simonsarris I agree that just eating pasture raised cows is not that much more complex, it's just weird to focus on the people who are avoiding the really bad thing and also avoiding some not-really-bad things, instead the people who are not even avoiding the really bad thing
@simonsarris this is a very "missing moods" kind of post, from my perspective. it is weird to focus on how actually not all instances of the class are bad given that very few people restrict themselves to only the non-bad instances of the class without cutting it out entirely
@simonsarris beef cows are not the only animals we eat; most chickens and (gestational) pigs live in horrifying conditions
vegetarians make a choice which avoids that suffering
it's true they could do a more complex thing which avoids most of it
but the suffering is bad!
@dundereloise Quality of their friendships
I am lucky to now have some good male friends but it seems easier to maintain friendships for women, for whatever reason
@IncidentNoodle@NathanpmYoung@Kirsten3531 my no-shows are usually bespoke, not correlated - someone got sick, car broke, baby didn't sleep, just an off day, whatever
if they all no-show because they got invited to the same other better party, sure; ime this is not common
@scottnarmstrong Video is https://t.co/sPfJkTMHRp, don't think the content is available in non-video form but the gist of it is to give the LLM Mathlib's style guide and walk through the relevant parts piece-by-piece asking the LLM how it can be improved.
@scottnarmstrong Are you planning on trying to upstream any of the foundational stuff into Mathlib? Tao has a good video where he talks about using LLMs for cleaning up Lean towards this end.
@underverse@KelseyTuoc@bigblackjacobin When he was writing, Anthropic's flagship model was Opus 3, which cost $75 per million output tokens. Fable (which is a wildly superior model) costs $50 per million output tokens.
Did my usual "bounce the drafts of the novels off a new model release"; everything prior has had a consistent set of misreadings of some things, but Fable cleared the deck of those, and also extracted all the most minor world stuff I buried obliquely so it could coo over it
@Tcho76521726@KelseyTuoc@totoTavrinky@shlevy "three strikes is a good policy, we just need a completely different set of laws, prosecutors, and judges first" is just not a useful analysis
@AiDeeply@littmath@pangram@signull yeah, a single instance of a claimed false positive is not what I would call "reliable reports" that it isn't accurate. even with a 1/10k false positive rate you would still expect to see that happening somewhere on the entire internet.
@AiDeeply@littmath@pangram pangram has some false negatives (you can make AI text which it will rate as human) but a vanishingly small false positive rate (if it says it's AI, you can pretty much trust that)
@appelbolt@tracewoodgrains@PradyuPrasad my impression was no but I don't follow that closely
I think this result would be pretty surprising if humans actually did add value; humans are generally not indistinguishable the way machines are
@appelbolt@tracewoodgrains@PradyuPrasad was there any game in the whole tournament (not counting the ones where the guy died) where any move played was not the top move selected by stockfish after 24 hours of runtime?
@tracewoodgrains@appelbolt@PradyuPrasad ICCF allows machines and it's consistently like a 10-way draw because no human adds any marginal value over stockfish
https://t.co/56E3Ju8YPh