NeurIPS paper: AI protein folding models converge on a shared two-stage mechanism
rishabh16_ · x · 2026-09-29
A Northeastern-led NeurIPS 2026 paper uses mechanistic interpretability on ESMFold, OpenFold, and Boltz-1, finding a shared two-stage computation: early blocks initialize pairwise biochemical signals like charge, late blocks build spatial distance/contact features. Steering these features induces predictable structural changes, and pairwise states are linearly interchangeable across models, suggesting convergent representational organization.
More from Research
- Calibrated Importance Sampling Fixes Training-Inference Mismatch in LLM RLVR — Tianrun Yu · 2026-09-29
- ZJU Releases EMem-Bench: 2,554 Episodes to Test Embodied Agent Memory — OmniAI-ZJU · 2026-09-29
- Tencent's AdaTutoRank Trains Setwise RAG Rerankers via Adaptive Tutoring Optimization — tencent · 2026-09-29
- Stanford lands $25M NIH grant to build AI center for personalized dementia care — StanfordAILab · 2026-09-29
- RL Gains May Hit a Wall: Verifiable Data Limits and Stalling nanochart Benchmarks — QuintinPope5 · 2026-09-29
- TALES Benchmark on LLM Game-Playing Accepted to NeurIPS 2026, New Results Coming — tw_killian · 2026-09-29