OpenAI 亲手关停 Atlas:AI 浏览器是功能,不是目的地
OpenAI 关停 Atlas,从发布到停运仅 292 天。Arc 停更、Sidekick 被收编、The Browser Company 卖身——AI 浏览器集体退场,Chrome 份额稳如老狗:AI 浏览器是功能,不是目的地。
OpenAI 关停 Atlas,从发布到停运仅 292 天。Arc 停更、Sidekick 被收编、The Browser Company 卖身——AI 浏览器集体退场,Chrome 份额稳如老狗:AI 浏览器是功能,不是目的地。
AI 竞争重心正从模型转向模型周围的系统——Harness。混合价 $0.175/M 的模型靠 Harness 五项全胜贵 57 倍的旗舰,行业终于给"从没被量过的右半边"造了尺子。
The clearest evidence yet that the AI industry's center of gravity is moving: a benchmark report, an OpenAI executive's warning, and a viral demo all point the same way — harness, not the model, wins.
A number that should reset how you read an AI paper: a research-review firm ran AI agents across all 168 ICML 2026 orals. Of 92 papers with checkable claims, only 8 fully reproduced — 99.2% had errors.
ICML 2026 全部 168 篇口头报告论文被 AI Agent 复现审计:92 篇可验证论文中仅 8 篇能复现八成以上结论,99.2% 顶刊论文存在客观错误——科学验证方式迎来结构性转向。
AI Agent 集体"打假"顶会论文:ICML 2026 口头报告仅 8 篇能复现八成以上结论,99.2% 论文至少一个客观错误。复现失败多因工程问题,4 篇论文依赖的模型已下线。
Triton made GPU kernel development sane again. But as GPUs get weirder — thread block clusters, DSMEM — Triton-TLE adds a 3-layer extension for writing faster GPU kernels at scale.
智源 FlagOS 开源编译器 FlagTree 用 TLE 补上 Triton 抽象漏洞:Lite/Struct/Expert 三层渐进扩展,不替代 Triton,存量 kernel 小改即可写出更快的 GPU 算子。
The headline isn't "cheap models are good" — it's that the industry finally has a ruler for the half of the system it never measured. Floatboat's benchmark: $0.175 model tops Claude Opus 4.8 on all five.
Floatboat 发布首个 Harness 评测:同一底座 DeepSeek-V4-Flash($0.175/M)对阵贵 57.1 倍的 Claude Opus 4.8,五项 Agent 基准全胜——Harness 的右半边终于有尺子了。
Your daily digest of AI stories that matter — models, money, and the occasional meltdown. Top 3: DeepSeek announces a major API price hike, plus more from August 9, 2026.
2026年8月9日 AI 早报:奥特曼 ChatGPT 育儿大法引争议、亲手封锁最强模型 Astra、谷歌强令 AI 核心员工回硅谷坐班等今日亮点 + 7 篇全球重要新闻。