Stanford's Kundaje: perturbation models barely generalize beyond training distribution
anshulkundaje · x · 2026-10-06
In a thread on genomic perturbation / virtual cell models, Stanford professor Anshul Kundaje argues that although SOTA perturbation models appear to predict effects of unseen perturbations in unseen conditions, this largely holds only for perturbations and conditions closely related to the training data and prior knowledge. That's fine, he says, but it must be acknowledged as a fundamental limit stemming from the learning strategy used.
More from Research
- User Sim Index is broken: trivial bot scores 95% across behavioral dims — ericzelikman · 2026-10-06
- Used OpenAI Dots as a Free Agent Swarm to Break a 47-Year-Old Math Record — jaxchang · 2026-10-06
- Stanford prof says current experiment designs can't train causal virtual cell models — anshulkundaje · 2026-10-06
- TIDES dataset on multi-party and multi-agent collaboration to debut at COLM 2026 — josephseering · 2026-10-06
- LakeQuest QA benchmark, testing RAG on messy enterprise tables and docs, hits COLM — hllo_wrld · 2026-10-06
- Frontier Data Summit lineup: Chollet, Dawn Song headline batch of new agent benchmarks — StanfordAILab · 2026-10-06