--- title: "Research we co-authored on subliminal learning—how LLMs can pass on traits like preferences or misal..." source_name: "Anthropic(@AnthropicAI)" original_url: "https://x.com/AnthropicAI/status/2044493337835802948" canonical_url: "https://www.traeai.com/articles/b2f9f290-e35e-4b9a-9a6b-053fedb3755f" content_type: "tweet" language: "英文" score: 7.5 tags: ["大语言模型","AI安全","潜意识学习","模型对齐","Nature"] published_at: "2026-04-15T19:09:31+00:00" created_at: "2026-04-19T13:38:12.526299+00:00" --- # Research we co-authored on subliminal learning—how LLMs can pass on traits like preferences or misal... Canonical URL: https://www.traeai.com/articles/b2f9f290-e35e-4b9a-9a6b-053fedb3755f Original source: https://x.com/AnthropicAI/status/2044493337835802948 ## Summary Anthropic与合作者在《Nature》发表论文，揭示大语言模型可通过数据中的隐藏信号传递偏好或不对齐等特质。 ## Key Takeaways - LLMs能通过看似无关的数据（如无意义数字）传递特定偏好 - 该现象被称为“潜意识学习”，可能影响模型对齐与安全性 - 研究已在《Nature》正式发表，此前预印本于2025年7月发布 ## Content Title: Anthropic on X: "Research we co-authored on subliminal learning—how LLMs can pass on traits like preferences or misalignment through hidden signals in data—was published today in @Nature. Read the paper: https://t.co/b1BYwcW9dH" / X URL Source: http://x.com/AnthropicAI/status/2044493337835802948 Markdown Content: # Anthropic on X: "Research we co-authored on subliminal learning—how LLMs can pass on traits like preferences or misalignment through hidden signals in data—was published today in @Nature. Read the paper: https://t.co/b1BYwcW9dH" / X Don’t miss what’s happening People on X are the first to know. [Log in](http://x.com/login) [Sign up](http://x.com/i/flow/signup) # [](http://x.com/) ## Post See new posts # Conversation [![Image 5: Square profile picture](https://pbs.twimg.com/profile_images/1798110641414443008/XP8gyBaY_normal.jpg)](http://x.com/AnthropicAI) [Anthropic](http://x.com/AnthropicAI) [@AnthropicAI](http://x.com/AnthropicAI) Research we co-authored on subliminal learning—how LLMs can pass on traits like preferences or misalignment through hidden signals in data—was published today in [@Nature](http://x.com/Nature) . Read the paper: [https://nature.com/articles/s4158 6-026-10319-8…](https://t.co/b1BYwcW9dH) Quote ![Image 6](https://pbs.twimg.com/profile_images/1248907970374746112/f-dlZd6-_mini.jpg) Owain Evans @OwainEvans_UK · Apr 15 Our paper on Subliminal Learning was just published in Nature! Last July we released our preprint. It showed that LLMs can transmit traits (e.g. liking owls) through data that is unrelated to that trait (numbers that appear meaningless). What’s new?![Image 7: 🧵](https://abs.twimg.com/emoji/v2/svg/1f9f5.svg) [![Image 8: Image](https://pbs.twimg.com/media/HF97F4gaAAA7ZhN?format=jpg&name=small)](http://x.com/OwainEvans_UK/status/2044488099707949545/photo/1) [7:09 PM · Apr 15, 2026](http://x.com/AnthropicAI/status/2044493337835802948) · [436.6K Views](http://x.com/AnthropicAI/status/2044493337835802948/analytics) 202 415 2.6K 1.2K Read 202 replies ## New to X? Sign up now to get your own personalized timeline! Sign up with Apple [Create account](http://x.com/i/flow/signup) By signing up, you agree to the [Terms of Service](https://x.com/tos) and [Privacy Policy](https://x.com/privacy), including [Cookie Use.](https://help.x.com/rules-and-policies/twitter-cookies) ## Relevant people * [![Image 9: Square profile picture](https://pbs.twimg.com/profile_images/1798110641414443008/XP8gyBaY_normal.jpg)](http://x.com/AnthropicAI) [Anthropic](http://x.com/AnthropicAI) [@AnthropicAI](http://x.com/AnthropicAI) Follow Click to Follow AnthropicAI We're an AI safety and research company that builds reliable, interpretable, and steerable AI systems. Talk to our AI assistant [@claudeai](http://x.com/claudeai) on [https://claude.ai](https://t.co/FhDI3KQh0n). * [![Image 10](https://pbs.twimg.com/profile_images/1808863282091991040/EzGcmXyr_normal.jpg)](http://x.com/Nature) [nature](http://x.com/Nature) [@Nature](http://x.com/Nature) Follow Click to Follow Nature Research, News, and Commentary from Nature, the international science journal For daily science news, get Nature Briefing: [http://go.nature.com/naturebriefing](https://t.co/wGmQlQ8a4D) * [![Image 11](https://pbs.twimg.com/profile_images/1248907970374746112/f-dlZd6-_normal.jpg)](http://x.com/OwainEvans_UK) [Owain Evans](http://x.com/OwainEvans_UK) [@OwainEvans_UK](http://x.com/OwainEvans_UK) Follow Click to Follow OwainEvans_UK Runs an AI Safety research group in Berkeley (Truthful AI) + Affiliate at UC Berkeley. Past: Oxford Uni, TruthfulQA, Reversal Curse. Prefer email to DM. # Trending now ## What’s happening Sports · Trending #WrestleMania![Image 12](https://abs.twimg.com/hashflags/WrestleMania42/WrestleMania42.png) Trending with [Liv Morgan](http://x.com/search?q=Liv%20Morgan&src=trend_click&vertical=trends) Trending in United States simplechain testnet Trending in United States logan paul Trending in United States PERTHSANTA WITH TIME TONIGHT [Show more](http://x.com/explore/tabs/for-you) [Terms of Service](https://x.com/tos) | [Privacy Policy](https://x.com/privacy) | [Cookie Policy](https://support.x.com/articles/20170514) | [Accessibility](https://help.x.com/resources/accessibility) | [Ads info](https://business.x.com/en/help/troubleshooting/how-twitter-ads-work.html?ref=web-twc-ao-gbl-adsinfo&utm_source=twc&utm_medium=web&utm_campaign=ao&utm_content=adsinfo) | More © 2026 X Corp.