AI News 2026-08-29 — Top 10 AI Stories

Curated AI news digest for 2026-08-29 (timezone Asia/Shanghai).

  1. An Anthropic researcher just gave us a peek at self-improving AI — TechCrunch AI. Given 10 benchmarks for specific misaligned behaviors, the automated systems were able to improve performance on every single one without degrading overall performance.
  2. Anthropic、Moderna先后引爆舆论场，AI制药到底该奖励什么？ — 雷锋网. 算法只是一小部分，但市场仍愿意为算法加价。 作者丨周蕾 编辑丨岑峰 上周，两条AI制药新闻在24小时之内，先后引燃了舆论场： 先是Anthropic发布早期药物研发成果：依托Claude大模型，他们几乎在无人工干预的情况下，自主设计出14个临床靶点的蛋白结合剂，全部实验数据公开，所合作的外部平台Twist Bioscience股价单日飙升22%。 次日，全球mRNA巨头Moderna与默沙东联合宣布，其个性化mRNA癌症疫苗intesmeran autogene在针对高危黑色素瘤的全球III期临床试验中达到主要终点，Moderna因此单日内股价翻倍。…
  3. Neocloud Lambda secures $1B in debt to buy more chips — TechCrunch AI. Neocloud Lambda has raised $1B in private debt to buy Nvidia AI chips and lease them to Microsoft. It's the latest in a string of loans, underscoring the high cost of the AI boom.
  4. Jalapeño 跑分炸场，GPU 推理路线开始分裂？ — 雷锋网. Token 成本成为大模型硬件竞争的新坐标。 作者丨郑佳美 编辑丨岑 峰 今天，OpenAI 的自研推理芯片 Jalapeño，终于从规格表走到了跑分台。 这次公开的数据很接地气：Token 吞吐、生成延迟、单位功耗下能跑多少推理。因为大模型上线以后，芯片面对的早就不只是一次训练任务。ChatGPT 要一直回消息，Reasoning 模型要持续生成长链路内容，Agent 还会一轮接一轮调用模型。算力再高，Token 吐得慢、延迟压不下来、功耗又高，数据中心照样难受。 成绩一出来，Reddit 很快就把 Jalapeño 和 NVIDIA Rubin 摆…
  5. Sora 为什么输给 Codex？ — 雷锋网. GPU 时间能否被复用，开始决定产品的扩张速度。 作者丨郑佳美 编辑丨岑 峰 8 月 23 日，奥特曼在 David Senra 的播客中谈到 OpenAI 内部的资源取舍时主动提到了 Sora。 他说得很直接：Sora 本身是个好产品，继续做下去也能成为一门不错的生意，但它实在太吃 compute，同一时期，Codex 的优先级更高，于是算力和团队投入开始往 Codex 倾斜。 但有趣的是，Codex 其实也一点都不省 GPU。Sora 生成一段视频，要让巨大的时空 latent 经过多轮 Transformer 计算；Codex 接到一句“把这个…
  6. Our decision on Cursor following its acquisition by SpaceX — OpenAI News. Our decision to wind down our contract providing OpenAI models to Cursor following its acquisition by SpaceX.
  7. Anthropic gets its first court win over the Pentagon’s supply-chain risk label — TechCrunch AI. A federal judge ruled the Trump administration illegally labeled Anthropic a supply-chain risk, handing the AI company a victory as its second Pentagon lawsuit continues in Washington.
  8. Supporting Thailand’s next generation of AI startups — OpenAI News. OpenAI and Thailand’s MHESI launch an eight-week accelerator helping 10 health, wellness, and education startups turn AI prototypes into trusted products.
  9. This Is How Anthropic Thinks AI Agents Should Navigate the Physical World — WIRED AI. The potential for AI to automate scientific research and manufacturing must be balanced with new risks, Anthropic says.
  10. 智谱 GLM-5.3-Flash上线，商汤大装置提供国产算力支持 — 量子位. 国产异构助力前沿智能进入普惠时代

Back to latest AI News