iiinova — where your digital self keeps evolving

Ulyana Piterbarg

@upiter.bsky.social

1.6K followers 330 following 5 posts

senior research scientist, Gemini upiterbarg.github.io

Ulyana Piterbarg @upiter.bsky.social · 12/02/2025

LMs trained to synthesize programs by repeatedly editing their own generations produce more diverse code compared to baselines This improves the trade-off between test-time FLOPs and pass@k

110

Ulyana Piterbarg @upiter.bsky.social · 12/02/2025

Our approach introduces an algorithm, LintSeq, for sampling across interdependent lines in source code by using a code linter With LintSeq, we can generate plausible edit *trajectories* for any source code file, covering possible ways of synthesizing its contents edit-by-edit with no linter errors

110

Ulyana Piterbarg @upiter.bsky.social · 12/02/2025

Our paper showing that LMs benefit from human-like abstractions for code synthesis was accepted to ICLR! 🇸🇬 We show that order matters in code gen. -- casting code synthesis as a sequential edit problem by preprocessing examples in SFT data improves LM test-time scaling laws

1102