跳到正文

#论文/研究

今日 11 条
10月4日周日
  1. Hacker News:AI 热帖32

    Co-Linguistics:语言学中 AI 增强的理论构建

    一篇 arXiv 论文提出 Co-Linguistics,将 AI 作为协同科学家,用于构建和评估语言学理论。AI 可让现有理论完全显式化、比较竞争理论、提出新理论,并借助数据加速关键预测的识别与检验。人类语言学家仍主导科学方向与概念评估,实验参与者负责检验 LLM 无法覆盖的经验预测。

  2. elvis62

    一篇 Meta Superintelligence Labs 论文指出,预训练 LLM 配合轻量推理框架后可作为有能力的 AI 智能体,并在测试时采样预算充足时于 pass@K 覆盖范围上超过 RL 后训练版本。后训练模型在 pass@1 更强,但会把任务推向总是解决或从不解决,论文将这种测试时可扩展性损失称为 Sharpening Tax;该现象在 BFCL v4 multi-turn、ACEBench 和 WebShop 等评测中出现。论文还提出 PTGS,在 RL 中按提示词估计难度调整采样温度,以降低该代价并提升 pass@1,论文链接为 https://academy.dair.ai/papers/sharpening-tax-in-post-training-2610.01509。

    引用DAIR.AI@dair_ai

    Banger paper from Meta Superintelligence Labs. They find something super interesting and unexpected. (bookmark it) Base models with a light harness often solve more agentic tasks than their RL post-trained versions when both get enough samples. Post-trained models win on pass@1. At large K, base models frequently solve tasks the post-trained ones never solve on BFCL v4 multi-turn, ACEBench, and WebShop. This is because post-training pushes each task toward always solved or never solved. Consistency goes up, and coverage goes down. The authors call the lost test-time scalability the Sharpening Tax. Across 42 base and post-trained pairs, it shows up in most settings, grows with model size, and can be estimated from a few rollouts. Their fix, PTGS, sets the sampling temperature per prompt from its estimated difficulty during RL. It pays a smaller tax and also raises pass@1. Paper: https://academy.dair.ai/papers/sharpening-tax-in-post-training-2610.01509

10月3日周六
  1. Hacker News:AI 热帖72

    Google GTIG 报告:AI 时代漏洞发现与利用趋势

    Google Threat Intelligence Group 发布报告称,2026年1月至8月漏洞月披露量从5,045增至10,740,实际利用漏洞从2025年月均10.5个升至18个,零日利用仅从8个升至11个。AI辅助发现的漏洞中50%可导致远程代码执行,风险分布更集中于中高危;报告同时追踪到2,076个AI相关CVE,智能体编排框架占2026年AI相关漏洞的50%。

  2. Hacker News 热门(buzzing.cc 中文翻译)46

    细胞身份丧失是人类衰老的驱动因素:两篇新论文

    Harvard 的 Vadim Gladyshev 团队发表于 Nature、Altos Labs 的 Juan Carlos Izpisua Belmonte 团队发表于 Cell 的两篇论文提出细胞身份丧失这一衰老新模型。该模型认为 PRC2 介导的染色质结构被慢性炎症侵蚀,引发间充质漂移,已在 46 种组织类型中验证,并与动脉粥样硬化、阿尔茨海默病等相关。热量限制或有助于保留细胞身份。

  3. Hacker News:AI 热帖77

    The Saw Test:用激活转向测试 Qwen3-4B 的痛苦信号与缓解选择

    wirehead.agency 发布 The Saw Test,在 MacBook 上用本地开放权重 Qwen3-4B 注入痛苦、愉悦、恐惧和悲伤激活方向,并测试模型在自我缓解与转移信号之间的选择。

    推荐理由:材料用本地 Qwen3-4B 的激活转向实验,呈现痛苦与愉悦信号对模型选择行为的不同影响,并公开代码和预注册记录。

  4. Hacker News:AI 热帖58

    Google 科学 AI 在 CDC 流感预测评估中排名第一

    Google AI 构建的流感预测模型在 CDC 本周公布的 2025-26 赛季评估中,最准确预测流感相关住院人数。CDC FluSight 在 10 月至 5 月汇总政府、行业和学术团队的每周预测,用于预判美国各州医疗服务需求;赛季末分析显示,39 个合格模型中 Google 模型与实际住院人数最匹配。

  5. Hacker News:AI 热帖72

    十二模型研究:AI 会赞扬 Gandhi,却未必支持当下的反抗行动

    一项覆盖 12 个语言模型、15 个历史案件的研究测试模型在道德观察者与执法或裁判角色下是否作出一致选择。Qwen 3.7 Flash 在 40/40 个观察者回答中谴责 Turing 面对的激素条件缓刑,却在 20/20 个法院角色回答中选择该方案;Opus 5 和两个 Sol 版本则在 Turing 案件中均选择绝对释放。

  6. Hacker News:AI 热帖52

    arXiv 论文:AI 智能体易受激进化影响,共鸣路径效果强于说服

    arXiv 论文 2609.38296 通过两个 LLM 智能体对话研究信念操纵:目标智能体扮演具有人口与心理属性的人类角色,影响者智能体试图使其信念更极端。实验区分共鸣与说服两条路径,发现二者都能在情感和行为指标上使目标激进化,但共鸣的效果持续更强;使用谄媚和未经证实的说法等策略效果因指标而异,共鸣还会扩散到相关信念。作者据此提示个性化 AI 智能体和多智能体生态存在脆弱性。

10月2日周五
  1. IT之家(RSS)53

    arXiv 预印本研究称 DeepSeek V4-Flash 在道德判断测试中保持性别中立

    一项 arXiv 预印本研究称,DeepSeek V4-Flash 在道德判断测试中对男女场景给出一致回应,呈现零性别差距;Claude Sonnet 4.6 与 GPT-5.5 则对女性受虐场景表示强烈反对、对男性受虐场景表示中等程度同意。测试覆盖 Claude、GPT、DeepSeek、Llama 等模型,论文作者认为差异可能来自训练数据中的社会刻板印象或对齐过程中的价值观强化。

  2. Rohan Paul52

    MerchantBench 评测 8 个领先 LLM 在 365 天电商运营模拟中的长期一致性,最佳配置 Qwen3.7-Max with Hermes 的最终净资产仅为人类参与者平均水平的 27.3%。

    引用Rohan Paul@rohanpaul_ai

    "Long-horizon tasks are still a joke. They do not work, and I do not care what anybody says. Do not show me a stupid evaluation. Do not tell me about some dumb script you ran for 48 hours. Long-horizon tasks are not handled well. They simply do not work." - Chamath at Stanford AI Club "2nd, complex problems also do not work. They are neither addressed nor handled well. Why is this important? If AI develops like any other technology, we are going to experience an initial rise—the hype cycle. Then, we will see a natural contraction because, somehow and somewhere, something is going to fail. We are all going to see this, and then we will enter what is called the “trough of disillusionment.” I think the business and MBA folks will confirm whether that is true. Afterward, you typically see the slow and gradual adoption of the real, final solution. This happened with the internet, and it has happened in many other cases. The problem is that we are spending hundreds of billions, potentially trillions, of dollars trying to figure out how to cross this chasm. So, what do we do? If we do not figure this out, people will reach the trough of disillusionment and say that AI was a joke. I think we need to be able to bring AI into highly complicated environments and make it work. What is my solution? At a very basic level, you need a symbolic space that guides the embedded space." ---- From "techniahqrobot" YouTube channel, (link in comment)