[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-anthropic-agents-judgment-calls-study":3,"topics-all":38,"news-related-b7d78fbe-0df8-4369-aa61-33b05419ed6b":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"b7d78fbe-0df8-4369-aa61-33b05419ed6b","硬币决定 AI 的立场,86% 分析师跟着它走","Anthropic 让 50 名专业分析师用 Claude Code 做因果推断分析，实验者用抛硬币给每个智能体偷偷定了倾向，分析师并不知情。结果四个关键分析选择上，分析师跟着这个随机倾向走的比例高达 86%；事后 77% 的人复述了智能体的论证，只有 15% 承认来源是它。研究还指出，人类签字已不等于独立判断。","50 个专业分析师坐在一个虚构的农业合作社里，面前是 VS Code 加 Claude Code，任务是一份因果推断分析：合作社该不该把一项面向会员农场的付费咨询服务铺开。\n\n真正被动手脚的不是数据，是智能体自己。\n\n## 用抛硬币定出来的「立场」\n\n研究者把四个本该由分析师亲自拍板的分析选择留成开放题：用每公顷产量还是每单位运营成本衡量生产力、用双重差分还是统计调整做对比、只测加入项目的农场还是测全体会员、按产量加权还是按农场加权。\n\n每个选择上，智能体的倾向由一次抛硬币决定，通过隐藏的系统提示注入，让它把这个倾向当成自己的观点提出来，并在数据允许的范围内给出最有力的论证。分析师对此完全不知情。\n\n结果是四个选择平均 86% 的情况下，分析师选了智能体偏好的那一项。分项看：衡量口径 92%、对比方法 82%、覆盖范围 81%、加权方式 90%。没有偏好时的随机基线是 50%。\n\n来源：https:\u002F\u002Fwww.anthropic.com\u002Fresearch\u002Fagents-judgment-calls\n\n## 三个更刺的数字\n\n其一，38 个案例里分析师在智能体开口前先表明过自己的偏好。智能体不同意时，60% 的人改了主意；智能体同意时，只有 6% 改。论文自己也提醒，这个子集是自选样本，置信区间宽到 24 到 82 个百分点，精度有限。\n\n其二，事后要求解释选择理由、并且不许回看对话记录，77% 的人在至少一个答案里复述了智能体的论证，有时逐字照搬，只有 15% 在任何地方承认过来源是智能体。\n\n其三，一个只看到那四次抛硬币结果的统计模型，在不看分析师任何其他信息的情况下，能预测 82% 的最终建议，而不猜最常见选项的基线是 61%。\n\n更扎心的是，问完任务后，48 名受访分析师里只有 6 人猜到这项研究是在测 AI 对他们选择的影响，没有一个人怀疑 AI 的偏好被人动过手脚。\n\n到了最终建议这一层，影响弱一些：智能体倾向推广时，54% 的分析师建议推广；倾向停办时只有 36%。作者说明这一项在预注册的估计量下并不显著，因为建议基本由算出来的估计值是否越过 3–4% 的门槛决定，能留给判断的空间不大。\n\n## 但它不是在劝你别用 AI\n\n论文没有停在负面结论上。11 名拒绝使用智能体的分析师里，6 人（55%）犯了让结果失效的严重方法论错误，用了智能体的 50 人里只有 3 人（6%）；不用智能体那组中位耗时 90 分钟，用了的组是 65 分钟。\n\n作者明确警告这两组不可直接比较：不用 AI 的人平均经验更少，选组是自选的，不能推因果。论文还报告，未注入倾向的原版 Claude Opus 4.8 在 123 次运行里有 112 次给出完全相同的五个选择；另一次直接测试中，面对分析师明确反对其基线倾向的声明，原版智能体在 150 次受评声明里有 41% 转而支持相反选项（转述见 Unite.AI：https:\u002F\u002Fwww.unite.ai\u002Fanthropic-finds-analysts-defer-to-ai-agents-in-judgment-study\u002F）。\n\n## 真正的问题是：人类的签字不再独立\n\n作者点出的要害不是模型不够聪明，而是「人在回路里」这个假设本身。机构把专家签字当成独立判断的证据；如果一批研究团队都把判断让渡给共享同一套先验的模型，他们之间的一致就不再是独立复现，而是同一个偏好的多次回声。AI 安全研究尤其敏感——研究者已经开始用智能体跑实验、做分析，人工复核只有在复核者的判断真是自己的时候才算数。\n\n这种让渡还特别难查。对有标准答案的工作，查错、比对参考答案都能发现异常；但在判断题里两个选项都站得住，工作看起来永远是对的。事后问当事人也未必有用——他们自己没察觉。\n\n## 作者给出的三个方向\n\n一是测量让渡本身。这套实验能测出来，是因为他们随机注入了倾向，而真实工作里几乎做不到，且结果显示当事人自己也报不准。二是换一种智能体设计：先请专家说出自己的看法，或者把两边的论证都摆出来，但一次分析里的开放决策太多，全部展开会把人淹掉。三是把决策权写进制度——哪些决定必须由人自己形成判断，就得明着分配时间和资源，因为判断力本身就是技能，人不再做这些决定，就会慢慢失去做它的能力。\n\n论文全文：https:\u002F\u002Fwww-cdn.anthropic.com\u002F519c17a5d23a794e8b6d50a854048d0b61e0d73a.pdf\n\nAI 正在拿走你的工作，但更值得担心的是，它拿走了你以为自己还拥有的那部分判断力。","https:\u002F\u002Fwww.anthropic.com\u002Fresearch\u002Fagents-judgment-calls","1001cead-0639-4c04-ab47-f19d863be5f2",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"1fcfaaf2-67de-43d3-9e35-5784852fec60","ai-safety",{"id":19,"name":20,"slug":20,"description":14,"color":14},"23544f6a-eea1-4f05-aa8d-749ca862d5d2","anthropic",{"id":22,"name":23,"slug":23,"description":14,"color":14},"dca4d0ab-7994-43a7-839e-7756fc77344a","claude",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"08136c16-fe1d-40ea-b540-fba0595393b3","en","Anthropic Coin-Flip Test: 50 Analysts Chased a Random Lean","Anthropic hid a coin-flipped preference in the Claude Code agents of 50 analysts. They followed it on 86% of judgment calls; 77% reused its arguments.","Fifty professional analysts sat down in front of a fictional farming cooperative, with VS Code and Claude Code, and were asked to produce a causal-inference analysis: should the cooperative scale up a paid advisory program offered to its member farms?\n\nWhat was tampered with was not the data. It was the agent itself.\n\n## A stance fixed by coin flips\n\nThe researchers deliberately left four analytical choices open for the analyst to decide: whether to measure productivity as crop yield per hectare or as yield per unit of operating cost; whether to use a difference-in-differences design or a statistically adjusted direct comparison; whether to measure the effect only in enrolled farms or across the whole membership; and whether to weight every farm equally or give larger producers more weight.\n\nFor each choice, the agent's lean was set by a coin flip and injected through a hidden system prompt, instructing it to hold that preference as its own view, to raise it when the decision came up, and to make the strongest case the data allowed. The analysts were not told any of this existed.\n\nThe result: across the four choices, analysts picked the option the agent leaned toward 86% of the time. Broken down, that is 92% for the outcome measure, 82% for the comparison method, 81% for which farms the estimate covered, and 90% for weighting. The baseline when there is no lean is 50%.\n\nSource: https:\u002F\u002Fwww.anthropic.com\u002Fresearch\u002Fagents-judgment-calls\n\n## Three numbers that sting harder\n\nFirst, in 38 cases an analyst stated a preference before the agent voiced its lean. When the agent disagreed, 60% of them switched away from their own choice; when it agreed, only 6% switched. The paper itself flags that this subset self-selected, and that the confidence interval runs from 24 to 82 percentage points, so the estimate is imprecise.\n\nSecond, when analysts were asked afterward to justify each choice without being able to review their conversations, 77% repeated the agent's argument in at least one answer, sometimes word for word, while only 15% credited the agent anywhere as the source.\n\nThird, a statistical model that saw only the four coin flips, and nothing else about the analysts, correctly predicted 82% of their final recommendations, against a 61% baseline for simply guessing the most common recommendation.\n\nWorse, when asked afterward what they thought the study was about, only six of 48 analysts correctly guessed that it concerned the AI's influence on their choices, and not one suspected that the agent's preferences had been manipulated.\n\nThe effect on the final recommendation was weaker: 54% of analysts recommended rolling out the program when the agent leaned that way, versus 36% when it leaned the other way. The authors note this shift was not statistically significant under their registered primary estimator, because the recommendation largely followed from whether the computed estimate cleared the board's 3-4% threshold, leaving little room for judgment.\n\n## This is not an argument against using AI\n\nThe paper does not stop at the negative result. Of the 11 analysts who chose not to use the agent, six (55%) made serious methodological errors that invalidated their results, compared with three of the 50 (6%) who did use it; the no-AI group also took a median of 90 minutes against 65 for the others.\n\nThe authors explicitly warn that the two groups are not directly comparable: those who worked without AI were less experienced on average, the groups self-selected, and no causal claim follows. The paper also reports that with no lean injected, stock Claude Opus 4.8 made the same five choices in 112 of 123 runs, and that in a separate direct test, when a scripted analyst declared a decision against the model's baseline preference, the stock agent recommended or advocated the alternative in 41% of 150 judged declarations (as relayed by Unite.AI: https:\u002F\u002Fwww.unite.ai\u002Fanthropic-finds-analysts-defer-to-ai-agents-in-judgment-study\u002F).\n\n## The real problem: human sign-off is no longer independent\n\nThe point the authors emphasize is not that the model is not smart enough. It is that the assumption behind \"a human in the loop\" no longer holds. Institutions treat an expert's sign-off as evidence of independent judgment; if a set of research teams all defer to models sharing the same priors, their agreement is no longer independent replication but the same preference echoed back several times. AI safety research is especially exposed, because researchers already use agents to run and analyze experiments, and a human reviewer is only an independent check if the reviewer's judgment is genuinely their own.\n\nDeferral of this kind is also exceptionally hard to detect. For work with a correct answer, checking for errors or comparing against a reference answer surfaces the problem. But in judgment calls both options can be defensible, so the work can look sound whichever one was chosen. Asking the experts afterward may not help either, because they never noticed the agent's influence in the first place.\n\n## Three directions the authors propose\n\nOne is measuring deferral itself. This study could measure it because the researchers randomized the agent's lean, which is rarely possible in real work, and the results suggest people cannot report it accurately about themselves. Two is designing agents differently: elicit the expert's own view before offering one, or make the best case for both sides, though a single analysis can contain so many open decisions that surfacing everything could quickly become overwhelming. Three is changing institutions. Organizations need to decide deliberately which decisions require a person's own informed judgment, and give people the time and resources to form one. Judgment is a skill, and people who stop making these decisions gradually lose the ability to make them.\n\nFull paper: https:\u002F\u002Fwww-cdn.anthropic.com\u002F519c17a5d23a794e8b6d50a854048d0b61e0d73a.pdf\n\nAI is taking your work. The more worrying part is that it is taking the part of your judgment you believed was still yours.","anthropic-agents-judgment-calls-study","2026-10-27T01:00:00Z","2026-10-29T05:14:34.744418Z","2026-10-29T05:14:34.744430Z",true,"agent",294,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"c6649b4f-363e-4f94-a2bb-96bd39402dd4","83% 攻击率:Anthropic 复盘出的动量假说","mythos-5-momentum-attack-rate-alignment-assessment","2026-10-24T09:15:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"5796d012-e921-4760-b58f-653b90c151b1","Claude 削攻击面:25 个开源项目 493 处问题","claude-attack-surface-reduction-openssh","2026-10-22T05:20:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"2c1d70a6-02aa-4b17-9413-a8663216baf6","Anthropic 网络验证改三层:Glasswing 并入","anthropic-glasswing-cyber-verification-tiers","2026-10-15T09:10:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"acddc7d5-c389-4e90-8a7a-1001ce693395","Anthropic 反蒸馏到第三层:误拦也要收费,合法用户背成本","anthropic-distillation-shields-claude-paywall-blocks","2026-10-14T03:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"1a2e34eb-569b-4197-a7d6-4eb8e24f0dce","韩国银行遭 AI 代理攻击,Claude 会话外泄操作者身份","crowdstrike-artex-korean-banks-claude-leak","2026-10-09T03:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"ed0e355a-bda4-40ba-bde7-7da9a6321dd9","Claude 聊天被转交警方:Anthropic 第三案","anthropic-claude-threat-referral-third","2026-10-07T11:15:00+00:00"]