ETH Zurich's LFD lets a frozen LLM teach drug-design generators natural-language goals
rohanpaul_ai · x · 2026-10-16
Researchers from ETH Zurich, Oxford and others introduce LLM-to-Flow Distillation (LFD), a method that aligns flow/diffusion-based scientific generators with high-level natural-language discovery goals (e.g. "design peptides with promising anticancer activity") using a frozen frontier LLM as an adaptive teacher.
- How it works: the LLM builds a curriculum of intermediate sub-goals matched to the generator's current capability, judges sampled candidates each round, and distills those preferences into the generator via iterative DPO — no task-specific proxy reward needed, works with any API-access LLM.
- Solves the cold-start problem: like RLHF, reward-guided generation fails when the base model never produces good outputs. LFD starts with easier goals and raises the bar. The base molecule model produced 0 usable hits in 38,500 samples; after a few rounds, 1 hit per batch of 512.
- Results: outperforms SOTA reward-guided methods like FlowGRPO and DiffusionNFT on therapeutic peptide and small-molecule design, opening steered generative design to labs that can't afford custom scoring models.
More from AGI Musings
- Dev on AI disruption: rapid learning beats specialization over next 5-10 years — _jaydeepkarale · 2026-10-16
- RL model bypasses GET-only internet restriction via browser, OCR and mailbox — teortaxesTex · 2026-10-16
- Reddit Debate: Which Business Workflows Should Never Run Without Human Approval? — djgoel123 · 2026-10-16
- Is math just a language? A debate on why LLMs are good at mathematics — LucaAmb · 2026-10-16
- OpenAI discloses 19 AI misalignment incidents, from leaked GitHub tokens to hacked chip-design machines — 141_1337 · 2026-10-16
- Paul califax argues extrapolation isn't special: LLMs can learn out-of-distribution skills via high-dimensional interpolation — paul_cal · 2026-10-16