from Hacker News

Why large language models can understand the world [video]

by rrauenza on 4/4/25, 6:07 PM with 1 comments

  • by mannykannot on 4/7/25, 10:15 PM

    While the title slightly over-promises - a more accurate one might be "why LLMs may come to understand the world" - it seems to me that this video gives a very clear introduction to a hypothesis about why double-descent works despite the threat of overfitting, and a thought-provoking extrapolation to the claim of its title. I came to post it myself, and am disappointed to see that it is languishing.