I worry that most of the focus on "AI for science" is getting it backwards. Try to do great science and build awesome technology. Build AI tools when it would be useful. Setting out to do "AI for science" will lead to all sorts of lamplight effects and square peg round holes.

the biggest lesson I've learned from the last few years is that the "tiny gap between village idiot and Einstein" chart was completely wrong

very interesting post tokenizers delenda est

is it possible to convert the ELO scale to centipawns? eg a difference of 500 ELO means you are 50-50 to win without your queen

just finished reading The 7 1/2 Deaths of Evelyn Hardcastle, a locked-room ...time-loop murder mystery extremely fun stuff it's surprising how much horror rests on the unknown

can any of my followers explain this?

TIL Picard is modelled on Hornblower! cc Joan of Arc Fan Account

idly thinking that a really fun and effective method of tutoring might be Step 1: Have student complete Math Academy Step 2: Ask a continuous series of "Why?" questions

I recently explored options trading semi-seriously and it has given me new insights into the nature of time. If you have ever bought or sold CALLS or PUTS, what's the most important or interesting thing you learned from it? (If a follower could reply smartly, please retweet.)

this is truly an incredible way to start your undergrad thesis

>Norman Borlaug saving literally ~a billion lives with genetic material from the original collection. wait, WHAT? i did not know this part