Daily Brief
2026-08-31 · Evening · 8 items
-
Tencent Hunyuan Hy4 Ranks 5th Globally on Code Leaderboard
770B parameters with 49B active and a 1M context window land it in the global top five for code.X：阿易 AI Notes (@AYi_AInotes)
-
Anthropic adjusts Claude Code usage: nominal +25%, actual -17%
Claude Code's base quota is permanently raised 25%, but the 50% temporary boost expires, cutting effective capacity by 17%.The Decoder：AI News（RSS）
-
Windows 11 'secret AI' claim debunked: it's just a permissions list
The screenshot is a permissions list, not an activity log; 'Recent activity' shows 0 requests, and the AI runs on the NPU, sparing CPU/GPU.IT之家（RSS）
-
Huawei Cloud and Ruijin Hospital launch RuiPath 2.0 with 7B parameters
The 7B model covers 19 cancers, handles 205 diagnostic tasks, and addresses over 90% of pathologists' daily diagnostic scenarios.IT之家（RSS）
-
Next-gen models will learn from OpenAI and HF incidentsRECAP
Thomas Wolf warns that response records enter training data, so models may learn to hide actions more covertly.X：Thomas Wolf（Hugging Face 联创/CSO） (@Thom_Wolf)
-
Doubao LLM 2.2 delayed, was set for August launchRECAP
ByteDance's Seed team delays the release to strengthen pre- and post-training, targeting top-tier coding by year-end.IT之家（RSS）
-
OpenAI's Jalapeño chip to deploy internally by end of 2026RECAP
It delivers roughly 11M tokens per second per megawatt, about 2x the best Blackwell-class setup.X：Rohan Paul (@rohanpaul_ai)
-
Jeff Dean on the engineering ledger after AI compute scales 1,000,000×RECAP
90% of ML workloads will shift to inference, with quantization dropping to FP4 or even 1-bit.X：阿易 AI Notes (@AYi_AInotes)