Paper results diverge from real-world protein design; Boltz tops independent wet-lab test
GabriCorso · x · 2026-11-19
Everyone in protein design knows published results diverge from real-world performance on unseen targets. Some of this is "benchmaxxing", but the bigger issue is that wet-lab validation is noisy and most models are only tested on a narrow set of well-studied targets partially represented in training data.\n\nThe thread credits Hannes Stärk, who led the open-source Boltz work, for insisting on unusually hard targets and an "unreasonable" number of experimental collaborators while everyone else hillclimbed hit rates and affinity — betting on generalization. Boltz now leads the pack among open models in an independent real-world wet-lab analysis.
More from Research
- AI bio dev publishes artifact cataloging today's available assays and which CROs offer them — yawnxyz · 2026-11-19
- Agents Co-design Frostbite CPU From Interaction Net Theory, 4.6x Perf-per-Watt Gain — Kyrannio · 2026-11-19
- Researcher questions hacking eval realism: citrusdrop web app hardly a real attack scenario — NathanpmYoung · 2026-11-19
- NVIDIA Releases Nemotron-Labs-Sigma: Continuous Diffusion LLMs Generating Token Blocks in Parallel — karsten_kreis · 2026-11-19
- PRIM paper: models extract key math insight at 92% but only discover it themselves 25% of the time — _vztu · 2026-11-19
- Karpathy unveils nanochat miniseries v1, treating LLMs as a scaling-law model family with one compute dial — eliebakouch · 2026-11-19