LFD distills LLM knowledge into flow models for reward-free scientific design
kastnerkyle · x · 2026-10-16
Researchers introduce LLM-to-Flow Distillation (LFD), which adapts scientific generators to natural-language discovery goals like 'design peptides with anticancer activity' by iteratively running DPO on LLM-generated sub-goals. It needs no task-specific proxy rewards, works with any frontier LLM via API, and outperforms reward-guided SOTA like FlowGRPO and DiffusionNFT on held-out benchmarks.
More from Research
- Logit lens on a Beat Saber finetune reveals it reuses language grammar circuits — ctjlewis · 2026-10-16
- Handwired compiles plain JavaScript into an exact, fully differentiable neural net — ctjlewis · 2026-10-16
- 128×128 record stands: how intermediate-work experiments foreshadowed reasoning models — ctjlewis · 2026-10-16
- TeamFusion: preference-based AI proxies debate to preserve minority views in team decisions — burkov · 2026-10-16
- MIT's C-GMT gives humanoid robots whole-body compliance without sacrificing motion tracking — MIT_CSAIL · 2026-10-16
- SUNY Albany to offer Topological Deep Learning course, from Perceptrons to WL test — burny_tech · 2026-10-16