ACE: New Interpretability Method Extracts Shared Circuits Across LLM Trainings Up to 32B
CatAstro_Piyush · x · 2026-11-19
A new interpretability method called ACE, appearing at NeurIPS 2026, tackles a fundamental puzzle: train the same AI three times and you get identical behavior but different circuits — so is a circuit the algorithm, or just one implementation of it?
- ACE extracts what different wirings share, offering an explanation for why models grok.
- The shared representations can also be used to steer LLM behavior at scales up to 32B parameters.
- Details are laid out in a thread, making this a notable new entry in mechanistic interpretability research.
More from Research
- RhoAgent in Nature Biotech brings AI design from proteins to programmable functional RNA — EricTopol · 2026-11-19
- Designing a humanoid robot foot: heel strikes hit 1511N, whole foot weighs 117g — dee_hw · 2026-11-19
- Google's GUARD paper: frozen base model, self-editing agent harness gains +8.89 points — omarsar0 · 2026-11-19
- Kezek splits mono two-person conversations into per-speaker tracks for full-duplex dialogue training data — manifestro · 2026-11-19
- Elie Bakouch: current AI research benchmarks overfit and can't track RSI — eliebakouch · 2026-11-19
- Hutter Prize text compression jumps ~10% by overfitting small transformers for arithmetic coding — ricklamers · 2026-11-19