Skip to content

Comment on Attention Is Bayesian Inference

Comments

Pretty interesting. The posterior matching is a big deal, but I'm not convinced by the handwaiving required to demonstrate it in larger models. I'm interested in seeing how direct EM training scales though.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.