Reproduce “LLM-to-Flow Distillation: Teaching Natural Language Goals to Scientific Generators” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Riccardo De Santi, Federico Di Gennaro, Angelo Gnazzo, Sophia Tang, Pranam Chatterjee, Fanny Yang, İsmail İlkan Ceylan, Michael M. Bronstein and 1 more
LFD adapts a flow generator to natural-language scientific goals by having a frozen LLM propose attainable sub-goals and judge sampled candidates, then training on positive-versus-negative preferences. It repeats this process as the generator changes. In GSK3β ATP-pocket design, LFD reached a mean GNINA CNNaffinity of 6.77 ± 0.11 pK, above reward-guided baselines; this is computational docking evidence, not experimental validation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.