Twitxy Saved

Marc Finzi

Verified account This account is verified. Learn more

@m_finzi

Research scientist at OpenAI. Previously postdoc at CMU and PhD at NYU.
San Francisco mfinzi.github.io Joined January 2017
364 Following 2.9K Followers 189 Posts

Latest posts 5

Marc Finzi Verified account @m_finzi
I revel in the chaos of the optimal square packing solutions
They鈥檙e upsetting only if you have not internalized the existence of pseudorandom number generators.

Simple questions should have simple and non arbitrary seeming answers right? Not for observers with limited compute
3 replies 1 reposts 13 likes 1.6K views View on X
Marc Finzi Verified account @m_finzi
Why do larger language models generalize better?

In our new ICLR paper, we derive an interpretable generalization bound showing that compute-optimal LLMs provably generalize better with scale! 馃搫arxiv.org/abs/2504.15208
1/7馃У
arxiv.org Compute-Optimal LLMs Provably Generalize Better With Scale Why do larger language models generalize better? To investigate this question, we develop generalization bounds on the pretraining objective of large language models (LLMs) in the compute-optimal...
5 replies 34 reposts 139 likes 39K views View on X