Stanford's Kundaje challenges Arc Institute model: exploiting data structure, not learning biology
anshulkundaje · x · 2026-11-20
Stanford professor Anshul Kundaje questions an Arc Institute-related denoising solution, arguing it exploits the structure of training data and its relationship to test cell types rather than learning actual biology.
His core claim: any weighted-means approach by definition interpolates along the training set, so as long as the training set isn't infinite, test performance depends entirely on whether test samples fall within the training set's interpolation zone. A substantive methodological critique of generalization claims in generative models for biology.
More from Research
- MIT's Hiroshi Ishii shares Tangible Media Group's Creative Tension principles — ishii_mit · 2026-11-20
- Train a real LLM entirely in your browser with this interactive demo — morgymcg · 2026-11-20
- Ex-OpenAI Researcher: English Is the New ML Programming Language, Lean the Type System — lukaszkaiser · 2026-11-20
- S1CogBench probes System One decision models across 24 cognitive tasks — Dengzhe Hou · 2026-11-20
- AFFECT from KAIST makes world models dynamics-faithful via action effect matching — John Won · 2026-11-20
- Mind2Act World: a 180-task benchmark exposing robots' mind-action composition failures — Mind2Act-World · 2026-11-20