Hugging Face Daily Papers · August 13, 2026 · 4 min read

Mechanist: AI as a Scientific Instrument for Discovering the Mechanisms of Intelligence

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

Mechanist: Discover how AI works, then make it better.</p>\n","updatedAt":"2026-08-13T09:35:17.651Z","author":{"_id":"620b3bbb0668e435407c8d0a","avatarUrl":"/avatars/e0fccbb2577d76088e09f054c35cffbc.svg","fullname":"Ningyu Zhang","name":"Ningyu","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":52,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8310112953186035},"editors":["Ningyu"],"editorAvatarUrls":["/avatars/e0fccbb2577d76088e09f054c35cffbc.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.12036","authors":[{"_id":"6a7d89f342823931a1f17385","name":"Mengru Wang","hidden":false},{"_id":"6a7d89f342823931a1f17386","name":"Junfeng Fang","hidden":false},{"_id":"6a7d89f342823931a1f17387","name":"Shuofei Qiao","hidden":false},{"_id":"6a7d89f342823931a1f17388","name":"Zhenqian Xu","hidden":false},{"_id":"6a7d89f342823931a1f17389","name":"Haoming Xu","hidden":false},{"_id":"6a7d89f342823931a1f1738a","name":"Haoxiong Wang","hidden":false},{"_id":"6a7d89f342823931a1f1738b","name":"Shumin Deng","hidden":false},{"_id":"6a7d89f342823931a1f1738c","name":"Linyi Yang","hidden":false},{"_id":"6a7d89f342823931a1f1738d","name":"Zhixiang Cui","hidden":false},{"_id":"6a7d89f342823931a1f1738e","name":"Xin Xu","hidden":false},{"_id":"6a7d89f342823931a1f1738f","name":"Yunzhi Yao","hidden":false},{"_id":"6a7d89f342823931a1f17390","name":"Buqiang Xu","hidden":false},{"_id":"6a7d89f342823931a1f17391","name":"Fei Shen","hidden":false},{"_id":"6a7d89f342823931a1f17392","name":"Haozhe Luo","hidden":false},{"_id":"6a7d89f342823931a1f17393","name":"Yunxiang Wei","hidden":false},{"_id":"6a7d89f342823931a1f17394","name":"Ningyu Zhang","hidden":false},{"_id":"6a7d89f342823931a1f17395","name":"Julian McAuley","hidden":false},{"_id":"6a7d89f342823931a1f17396","name":"Tat Seng Chua","hidden":false},{"_id":"6a7d89f342823931a1f17397","name":"Huajun Chen","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/620b3bbb0668e435407c8d0a/1cvyGUe8OkEfiJvl4MJvm.mp4"],"publishedAt":"2026-08-12T00:00:00.000Z","submittedOnDailyAt":"2026-08-13T00:00:00.000Z","title":"Mechanist: AI as a Scientific Instrument for Discovering the Mechanisms of Intelligence","submittedOnDailyBy":{"_id":"620b3bbb0668e435407c8d0a","avatarUrl":"/avatars/e0fccbb2577d76088e09f054c35cffbc.svg","isPro":true,"fullname":"Ningyu Zhang","user":"Ningyu","type":"user","name":"Ningyu"},"summary":"AI models have achieved remarkable success across diverse domains, yet the mechanisms underlying their capabilities and the risks they may pose remain poorly understood. As AI development becomes faster and increasingly automated, mechanistic exploration remains largely manual, widening the gap between what models can do and our ability to understand and control them. To bridge this gap, we introduce Mechanist, an agentic system that uses AI as a scientific instrument for the autonomous discovery of mechanisms underlying AI intelligence. To support autonomous mechanistic discovery, we construct an interpretability-focused knowledge graph of approximately 13,000 papers and integrate it with a multidisciplinary database of 43 million papers spanning 26 fields. We further curate a library of 32 foundational methods for mechanism analysis, causal intervention, and validation. Compared with Claude Code and existing AI-scientist systems, Mechanist generates more valuable mechanism hypotheses and executes experiments more reliably. Mechanist also demonstrates a progression from discovering model behaviors to explaining and controlling AI models. Specifically, Mechanist first uncovers a counterintuitive safety risk in scientific laboratories, showing that unsafe traits can transfer across modalities through apparently safe training data. Mechanist then develops a mechanism theory of belief, revealing how models represent world knowledge, form beliefs, infer the beliefs of others, and how these mechanisms emerge during pretraining. Finally, Mechanist translates these mechanistic insights into practical interventions that improve model performance across diverse scenarios and steer scientific foundation models toward generating DNA sequences with specified properties.","upvotes":55,"discussionId":"6a7d89f442823931a1f17398","projectPage":"http://mechanist.openkg.cn/","githubRepo":"https://github.com/zjunlp/Mechanist","githubRepoAddedBy":"user","ai_summary":"Mechanist is an autonomous agentic system that uses AI to discover and control the mechanisms underlying model intelligence, generating hypotheses, performing causal interventions, and improving safety and performance.","ai_keywords":["mechanistic interpretability","causal intervention","knowledge graph","mechanism analysis","belief representation","cross-modal transfer","scientific foundation models"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":18,"organization":{"_id":"620a6fcd8d5e5dfed284bc91","name":"zjunlp","fullname":"ZJUNLP","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1644851027419-620a61cba53066560e226d30.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"620b3bbb0668e435407c8d0a","avatarUrl":"/avatars/e0fccbb2577d76088e09f054c35cffbc.svg","isPro":true,"fullname":"Ningyu Zhang","user":"Ningyu","type":"user"},{"_id":"64bf898d979949d2e2585c9a","avatarUrl":"/avatars/da77c856ec997e2b812c06272a01c8b2.svg","isPro":false,"fullname":"mengruwang","user":"mengru","type":"user"},{"_id":"64300415b009240418dac70c","avatarUrl":"/avatars/5175cdbc7683b0b52d5c742e93d3be83.svg","isPro":false,"fullname":"Qu Yang","user":"quyang22","type":"user"},{"_id":"6a1416c472ec40708cfcee2e","avatarUrl":"/avatars/a75a553b5d105bd39dddfe778690d263.svg","isPro":false,"fullname":"lilya","user":"lilya2026","type":"user"},{"_id":"6a1443f02a9759cfbdf80a48","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a1443f02a9759cfbdf80a48/dIoHGMzJ3U6k7VTOaBd4Y.jpeg","isPro":false,"fullname":"Haoxiong Wang","user":"WangHX2026","type":"user"},{"_id":"684bc1be17ae31ba66171292","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/684bc1be17ae31ba66171292/LFlkU4kArMjSzIbwjXd44.jpeg","isPro":false,"fullname":"Jingsheng Zheng","user":"JohnsonZheng03","type":"user"},{"_id":"66abc6da92b9eb71fe476118","avatarUrl":"/avatars/6d1618f45cc76da80335ad926ad24552.svg","isPro":false,"fullname":"xy.r","user":"ShawnRu","type":"user"},{"_id":"66cd5c0ad0bdbf5d712cab41","avatarUrl":"/avatars/29115c021c6ecaae889711d20a24febd.svg","isPro":false,"fullname":"YLR9933","user":"YLR9933","type":"user"},{"_id":"679e1f7c31bab0a2a309d61f","avatarUrl":"/avatars/116912ef6a154edec9d589e0e0597fc9.svg","isPro":false,"fullname":"Zhenqian","user":"ZhenqianXu","type":"user"},{"_id":"65535b54140fc44a74d43635","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/MIrD8OzDKF2aI38i7ZPjR.jpeg","isPro":false,"fullname":"Zhisong Qiu","user":"consultantQ","type":"user"},{"_id":"67e817ceba868546ac409f92","avatarUrl":"/avatars/83d829730eddabac0fee910f020583eb.svg","isPro":false,"fullname":"Liu Xinjie","user":"LiuXJ-kai","type":"user"},{"_id":"672c198760bdd070539fd7ed","avatarUrl":"/avatars/1064f0f5929c505589ee77f3e36df7e9.svg","isPro":false,"fullname":"Bohao Wang","user":"Baymax0110","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"620a6fcd8d5e5dfed284bc91","name":"zjunlp","fullname":"ZJUNLP","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1644851027419-620a61cba53066560e226d30.png"},"query":{}}">

Papers

arxiv:2608.12036

Mechanist: AI as a Scientific Instrument for Discovering the Mechanisms of Intelligence

Published on Aug 12

· Submitted by

Ningyu Zhang on Aug 13

ZJUNLP

Upvote

Authors:

Abstract

Mechanist is an autonomous agentic system that uses AI to discover and control the mechanisms underlying model intelligence, generating hypotheses, performing causal interventions, and improving safety and performance.

Generated by thinkingmachines/Inkling-Small

AI models have achieved remarkable success across diverse domains, yet the mechanisms underlying their capabilities and the risks they may pose remain poorly understood. As AI development becomes faster and increasingly automated, mechanistic exploration remains largely manual, widening the gap between what models can do and our ability to understand and control them. To bridge this gap, we introduce Mechanist, an agentic system that uses AI as a scientific instrument for the autonomous discovery of mechanisms underlying AI intelligence. To support autonomous mechanistic discovery, we construct an interpretability-focused knowledge graph of approximately 13,000 papers and integrate it with a multidisciplinary database of 43 million papers spanning 26 fields. We further curate a library of 32 foundational methods for mechanism analysis, causal intervention, and validation. Compared with Claude Code and existing AI-scientist systems, Mechanist generates more valuable mechanism hypotheses and executes experiments more reliably. Mechanist also demonstrates a progression from discovering model behaviors to explaining and controlling AI models. Specifically, Mechanist first uncovers a counterintuitive safety risk in scientific laboratories, showing that unsafe traits can transfer across modalities through apparently safe training data. Mechanist then develops a mechanism theory of belief, revealing how models represent world knowledge, form beliefs, infer the beliefs of others, and how these mechanisms emerge during pretraining. Finally, Mechanist translates these mechanistic insights into practical interventions that improve model performance across diverse scenarios and steer scientific foundation models toward generating DNA sequences with specified properties.

View arXiv page View PDF Project page GitHub 18 Add to collection