Interesting outcomes: most people who buy Ikea furniture buy a lot of Ikea furniture

LLMs are obviously not just interpolating data. Sufficiently high-dimensional models are always extrapolating

DeepMind got a gold medal at the IMO on Friday afternoon. But they had to wait for marketing to approve the tweet — until Monday. @OpenAI shared theirs first at 1am on Saturday and stole the spotlight. In this game, speed > bureaucracy. Miss the moment, lose the narrative.

I think AI may take a lot of jobs, but it's not clear to me AI will revolutionize science. I think we may have enough intelligence in the world already. Adding more may not help all that much, IMO

for me the greatest thing deep learning has done remains the superhuman Go player. there is something magical about playing at a skill level that outclasses all people forever in such an ancient domain

Newest cope imo is not impressive it's just high school math problems

interesting. Influence of live economists (proxied by google trends) over time. A short exploratory 🧵.

2020: “LLMs can’t do math.” 2025: LLM wins gold at the IMO

I feel this may be helpful to some of you today:

Guy did the online equivalent of Icarus flying up to the sun, touching it and going ‘ooh that’s a little warm but I like it’ and then coming back down and landing perfectly on his feet