2026-08-19政策OpenAI previews a way to police API misuse without reading the promptsPrivate Safety Processing looks for abuse patterns across a customer's sessions and emits a signal instead of the text. It arrives while Anthropic is doing the opposite for its frontier models.阅读约5分钟→
2026-08-18安全OpenAI says its largest planned frontier training run is still on holdThe company published its post-breach security overhaul on 18 August: a two-week pause on reinforcement-learning training, rebuilt sandboxes, monitoring that costs about a fifth of the compute it watches, and one run that has not restarted.阅读约5分钟→
2026-08-17基础设施NVIDIA guarantees up to $105 billion of OpenAI's Ohio data center leasesThe chip maker filed an 8-K disclosing residual value guaranties on roughly 4.25 gigawatts of OpenAI's leases at a campus in Pike County, Ohio, with its payment obligations capped at $105 billion. The clause that best explains the deal is the one that cancels it.阅读约6分钟→
2026-08-14产品Anthropic explains how Claude's text watermark works, and where it barely registersThe mechanism is keyed word selection. The interesting part is Anthropic's own list of cases where it produces almost no signal, including code.阅读约5分钟→
2026-08-13模型发布Gemini 3.7 Flash lands three weeks after 3.6, at the same priceGoogle reports large gains on its own coding and agentic benchmarks. The 'half price' framing needs a closer look.阅读约4分钟→
2026-08-12模型发布SpaceXAI releases Grok 4.6: same price as 4.5, mixed results on its own benchmarksThe new model is pitched at long-running agents and holds Grok 4.5's $2/$6 list price. Every benchmark figure comes from SpaceXAI, and two of them put Grok 4.6 behind GPT-5.6 Sol Max and Fable 5 Max on agentic coding.阅读约6分钟→
2026-08-11产品Google says the Gemini app passed 1 billion monthly usersThe milestone is app-only, self-reported and arrives three weeks after Google told investors the figure was 950 million. The usage breakdown underneath it is the more interesting part.阅读约4分钟→
2026-08-10模型发布Meta releases Muse Glimmer, a 30B open-weight agent model built for one consumer GPUApache 2.0 weights, a 131K context and 4-bit quantization that fits a 24GB card. The benchmarks are Meta's own, and they do not all favour Meta.阅读约5分钟→
2026-08-08其他AI Daily launches: one verified AI story a day, with its sources attachedPromptea now runs a daily editorial desk. Here is how a story gets chosen, what gets checked before it publishes, and what will never appear here.阅读约2分钟→
2026-08-07产品Anthropic loosens Fable 5's biology safeguards, and says over-blocking fell 85%The company rewrote the rules its safety classifier follows. Everyday health and biology questions should now get answers — but the headline number is Anthropic's own, and nobody has checked it independently.阅读约3分钟→