Stanford researcher clashes over genomics foundation model's cherry-picked variant prediction benchmarks
anshulkundaje · x · 2026-10-21
A debate over a new genomics foundation model: Anshul Kundaje notes it is not a pure DNA language model but does multimodal reconstruction, which raises causal-interpretation concerns without careful architecture design, though it's fine for prediction. Critic Dmitry Penzar asks why, if the model is truly better at variant prediction, the team only reports selected AG benchmarks instead of all of them, and avoids apples-to-oranges track prediction comparisons. Given the same team's earlier "revolutionary" Evo/Evo2 claims, he is skeptical of every number. Kundaje says the claims can be tested if the model is released.
More from Research
- PyTorch details KTIR and Dataflow Scheduler behind IBM's Spyre accelerator backend — PyTorch · 2026-10-21
- Anthropic: MLP Neurons Are a Stronger Basis for Interpretability Than Assumed — omarsar0 · 2026-10-21
- Microsoft's Self-HPO: Coding Agent Redesigns Its Own Harness, +7 Points on SWE-bench — mark_k · 2026-10-21
- Tim Dettmers: Academia is about to have a renaissance — the best AI work of the next decade happens in university labs — StasBekman · 2026-10-21
- Stanford and Laude release Atlantic-CC, a 17T-token model-filtered Common Crawl for open LLMs — Diyi_Yang · 2026-10-21
- AI saves scientists 7 hours a week, but validation is the new bottleneck — AnneliesGamble · 2026-10-21