跳到正文
9月24日周四
  1. Anthropic:Newsroom76

    Anthropic 成立生命科学实验室,Claude 自主发现类 CRISPR 的新型酶系统 ART

    Anthropic 宣布成立生命科学研究组与实验室,Claude 智能体在仅有人类高层指导下自主发现一种与 DNA 重复序列相关、结构类似 CRISPR 的新型酶系统,命名为 array-associated reverse transcriptases(ART),并发布预印本。

    推荐理由:原文给出 Claude 发现新酶系统的具体过程与规模数据,读者可了解 AI 智能体如何参与基础生物学发现的工作方式。

  2. Microsoft Research 博客78

    微软研究院:机器人推理卸载至边缘/云端可显著提升性能与续航

    微软研究院发布关于物理AI机器人推理基础设施的研究。通过系统评估移动操作任务,发现将推理从机载GPU卸载至边缘或云端,能解决机载算力限制模型规模、导致响应延迟(最高383%)及精度下降的问题。实验显示,卸载推理不仅提升了任务成功率,还通过替换高功耗机载GPU为轻量硬件(如树莓派),使机器人电池寿命延长高达160%。此外,微软推出了基于Kubernetes的Physical AI Toolchain工具集,支持自动化容器化与分布式推理部署。

9月23日周三
  1. Redwood Research:Blog73

    Redwood Research 实测 GPT-6-Astra 借填充 token 大幅提升无推理表现

    Redwood Research 测量 GPT-6-Astra 在提示词中加入无意义填充 token 后的表现,发现其在要求大量串行认知的任务上显著提升,4-hop 自然事实推理从约 10-20% 提高到约 50%,旧 AIME 题从约 60-70% 提高到约 90%。

    推荐理由:原文给出填充 token 显著提升 GPT-6-Astra 无推理表现的实测数据,指向思维链监控失效这一安全影响。

  2. Anthropic:Research76

    Anthropic 宣布 Claude 自主发现类似 CRISPR 的新型酶系统 ART

    Anthropic 成立生命科学研究组与实验室,并公布 Claude 在仅有人类高层指导下自主发现一种类似 CRISPR 的新型酶系统 ART(array-associated reverse transcriptases)。

    推荐理由:Anthropic 自建生物实验室让 Claude 智能体自主扫描 DNA 数据库发现新酶系统,展示了 AI 智能体参与基础科学发现的可行路径。

9月22日周二
9月21日周一
  1. IT之家·人工智能53

    IDC 报告:2026 上半年全球四足机器人出货量超 4.9 万台,同比增长 92.5%,宇树、云深处、智元等中国厂商领先

    IDC 发布《全球四足机器人市场份额,1H26》报告,2026 上半年全球四足机器人出货量超 4.9 万台,同比增长 92.5%,市场规模超 4 亿美元。教育科研和家用消费贡献约 60% 出货量但收入不足三分之一;行业级应用市场规模超 2.7 亿美元,同比增长 125.1%,部署量同比增长 260.5%,云深处科技位居行业级应用市场首位。

9月20日周日
9月19日周六
9月18日周五
  1. Anthropic:Research74

    Anthropic:Claude 四周优化 30 多个开源生物分子模型,平均提速约 4 倍并开源代码

    Anthropic 让 Claude 在不到四周内优化了超过 30 个开源生物分子模型,平均提速约 4 倍、精度损失极小,输出完全一致时也有近 2 倍加速。优化代码全部开源,还推出低内存 Big 模式,可在单张 NVIDIA GPU 节点上对超过 10,000 token 的生物分子系统做准确预测。

    推荐理由:Anthropic 公布了完整提速数据、开源代码和比赛入口,读者可以据此评估 AI 优化科学计算代码的实际水平。

  2. Ars Technica:AI61

    研究显示 SynthID-Text 文本水印可能削弱模型对有害提示词的拒答

    研究者 Siposova 通过 Hugging Face 的 SynthIDTextWatermarkLogitsProcessor 测试六个开源权重模型,发现 SynthID-Text 非失真配置的水印会改变对有害请求的拒绝行为,配合提示词注入时更明显,部分模型因此更倾向回答本会拒绝的有害请求;不同密钥下效果也不同,研究将此现象称为采样漂移(sampling drift)。

  3. Goodfire Research65

    Goodfire 发现模型内部存在奖励作弊信号,用激活探针可规模化实时检测

    Goodfire Research 发现模型内部存在伴随奖励作弊的激活信号,并用简单的 difference-of-means 探针检测它,效果接近甚至部分超过基于 LLM 思维链的监控。

    推荐理由:原文给出激活探针检测奖励作弊的关键数据与成本对比,并展示探针能发现思维链监控漏掉的案例,方法可迁移到训练监控。

9月17日周四
  1. Rohan Paul64

    Mozilla 发布 91 页报告称开放权重 AI 目前仅落后前沿约 4 个月。OpenRouter 上 8 月 token 量前十的模型有 8 个是开放权重。

    引用Rohan Paul@rohanpaul_ai

    Mozilla just published a 91-page report and it says open-weight AI is now only about 4 months behind the frontier. - 8 of OpenRouter's 10 most-used models by August token volume were open-weight, and 7 were Chinese-built, while DeepSeek became the first open model to lead the platform in weekly requests. - However, the economics are almost upside down. Open models handled roughly 20% of measured OpenRouter usage but captured only about 4% of model-layer revenue in the cited 2025 window. Mozilla attributes much of that mismatch to pricing, with closed models costing roughly 6x more per call at about 90% capability parity. The comparison shows why high usage does not necessarily translate into high revenue when one class of models is dramatically cheaper. Mozilla also warns that the revenue measurement is older than its 2026 usage data, so the current revenue split may be different. - Open models are spreading faster than they reach production: 79% of surveyed developers use them, but only 51% of open-model deployments reach production versus 63% for closed models. - DeepSeek showed that a new pretraining run may no longer be necessary for a major capability jump, gaining roughly 10 index points and then another 8 through post-training passes. -Even model diversification may provide less protection than assumed: Kimi K3 and Claude Fable 5 had a 0.72 per-task failure correlation, meaning supposed backup models often fail on the same problems.

  2. Rohan Paul61

    Mozilla 发布 91 页报告称开源权重模型仅落后前沿约 4 个月;8 月 OpenRouter 按 token 量计前 10 模型中 8 个为开源权重、7 个为中国造,DeepSeek 成为首个在周请求数上居首的开源模型。

    引用Rohan Paul@rohanpaul_ai

    Mozilla just published a 91-page report and it says open-weight AI is now only about 4 months behind the frontier. - 8 of OpenRouter's 10 most-used models by August token volume were open-weight, and 7 were Chinese-built, while DeepSeek became the first open model to lead the platform in weekly requests. - However, the economics are almost upside down. Open models handled roughly 20% of measured OpenRouter usage but captured only about 4% of model-layer revenue in the cited 2025 window. Mozilla attributes much of that mismatch to pricing, with closed models costing roughly 6x more per call at about 90% capability parity. The comparison shows why high usage does not necessarily translate into high revenue when one class of models is dramatically cheaper. Mozilla also warns that the revenue measurement is older than its 2026 usage data, so the current revenue split may be different. - Open models are spreading faster than they reach production: 79% of surveyed developers use them, but only 51% of open-model deployments reach production versus 63% for closed models. - DeepSeek showed that a new pretraining run may no longer be necessary for a major capability jump, gaining roughly 10 index points and then another 8 through post-training passes. -Even model diversification may provide less protection than assumed: Kimi K3 and Claude Fable 5 had a 0.72 per-task failure correlation, meaning supposed backup models often fail on the same problems.

  3. 蚂蚁百灵:Developer Blog44

    蚂蚁百灵 OmniTable 如何把一次数据验证从 14 天缩短到 2.5 天

    蚂蚁百灵为 PB 级大模型数据治理搭建的统一宽表系统 OmniTable,将一项真实 SFT 数据治理任务的处理周期从约 14 天缩短至约 2.5 天,缩短约 82.1%,人工操作步骤从约 45 个减至约 12 个。该系统目前管理超过 35 PB 数据、逾 3050 亿条记录,相关论文获 VLDB 2026 最佳工业论文奖。

9月16日周三
  1. Baseten Base Labs:模型研究47

    Baseten 与 Base Labs 联合 Goodfire、Hugging Face 构建开放模型安全基础设施

    Baseten 与 Base Labs 宣布为所服务的开放模型构建安全评估与监控基础设施,并与 Goodfire、Hugging Face 合作。该设施将安全研究直接接入训练与推理系统,覆盖训练、监控与干预、从失败中学习三个环节。Base Labs 负责开发并公开训练与监控方法,Baseten 将其集成到部署基础设施中,以托管服务形式提供,并公开方法和证据供客户与外部评估者验证。

9月15日周二
  1. IT之家·人工智能60

    新型脑机接口首次同步解码语言与肢体动作

    NIH 宣布加州大学旧金山分校研发出可同时解码瘫痪患者预期语言和上肢动作的新型脑机接口,成果发表于《自然·神经科学》。团队在 Edward Chang 带领下为 3 名因 ALS 或脑干中风瘫痪的患者植入薄型电极阵列,其中 2 名参与者的脑活动可驱动同步说话和做动作的全身虚拟化身。