YC Paper Club focuses on data: multilingual pre-training and agent benchmarking

ycombinator · x · 2026-08-21

Y Combinator announced this week's Paper Club theme is data, addressing how old assumptions about training and evaluation are breaking as models scale across tasks and languages. Four domain experts were gathered to present: Francois Chanut on the importance of data, Vincent Chen on the art and science of benchmarking agents, Valentin Kuleshov on diffusion language models for production, and Shayne Redford on ATLAS and practical scaling laws for multilingual models.

Related event: YC Paper Club spotlights data, benchmarks and multilingual models(3 posts)→

Original post →

More from Companies & People

Companies & People channel →