跳到正文
arXiv cs.AI· Piotr Jedryszek, Oliver M. Crook·· 2 小时前AI 评分25

走向 Boltz-1 稀疏自编码器的高效自动可解释性

Towards Efficient Auto-Interpretability of Boltz-1 Sparse Autoencoders

AI 导读

一项 arXiv 研究提出针对 Boltz-1 Pairformer trunk 稀疏自编码器(SAE)的自动可解释性流程,将潜在特征的一致性、可描述性和预测力整合为统一管线。

正文 · 原文

View PDF HTML (experimental)

Abstract:Sparse autoencoders (SAEs) could turn AI models in Biology and other fields into engines of scientific discovery by explaining the superhuman capabilities of those models. We postulate that a latent is useful if we know three things: whether it is coherent, whether it can be described, and whether that description has predictive power. These questions are routinely conflated. We assemble them into a single pipeline and report some practical innovations. First, cross-seed dictionary stability prioritises which latents are worth spending resources to investigate. Second, an intruder-detection task asks whether a latent's activating examples share a recognizable pattern. Third, a separate pass proposes a candidate biological description which we convert into falsifiable predictions which can be tested in silico. Deployed on the Boltz-1 Pairformer trunk, stability prioritisation finds interpretable latents using about 4.4 times fewer latent evaluations each, and at 5.2 times lower measured LLM cost, while recovering over half of them. However, we notice that in our setup the LLM judge is heavily biased towards consistent residue identity and a simple filter that selects for features with high amino-acid identity precision outperforms the stability thresholding. This simpler method has the added benefit of requiring no extra SAE seeds but it remains uncertain if it reflects a genuinely useful heuristic or an limitation of the LLM judge.
Subjects: Quantitative Methods (q-bio.QM); Artificial Intelligence (cs.AI)
Cite as: arXiv:2608.27754 [q-bio.QM]
  (or arXiv:2608.27754v2 [q-bio.QM] for this version)
  https://doi.org/10.48550/arXiv.2608.27754

arXiv-issued DOI via DataCite

Submission history

From: Piotr Jedryszek [view email]
[v1] Thu, 27 Aug 2026 22:32:43 UTC (169 KB)
[v2] Mon, 16 Nov 2026 21:33:26 UTC (206 KB)

来源:arXiv cs.AI · arxiv.org