AI — Page 7

OpenAI's November 5th compact covers future campuses it owns and operates, with a pledge not to require NDAs from public officials and a push for industry-wide rules.

Baidu published the 280B-parameter pre-training checkpoint on November 5th, 2026, with 14B active parameters, a 1M-token inference claim, and tighter custom terms than ERNIE 4.5.

The FT says Microsoft declined an option worth tens of billions, leaving Nscale to find new customers for the Texas expansion ahead of its planned IPO.

Muse's cloud browser has to sign in from scratch on sites that may demand one-time codes or block it outright. Muse 10.0 contains an "Add tab" flow, a gated Chrome extension row promising more reliable control, and local bridge code that could start from the Chrome tab a user has already authenticated, although RuntimeWire has not seen the flow work live.

Tencent released Hy4 on November 4th through hosted APIs and named products including TokenHub, WorkBuddy, CodeBuddy, Yuanbao and ima. Model weights and a system card remain pending.

ADWEEK reports Claude carried attributed AP election data on Election Night, days after Anthropic filed its public S-1 on October 26th.

The November 4th release brings GIFs, reusable reference images and editable frames to ChatGPT and Codex, plus two API models with 50% faster image generation and Flare output in as little as six seconds.

Samir Ahmed says the latest Dots release adds inline iOS replies, faster calls, a spoken hang-up and more reliable Windows MCP workflows.

Across 220 GDPval tasks, Gemma 4 31B used 23% of GLM 5.2's energy for 56% of its output value. Eight H100s reduced completion time by 22% while using 3.24 times the energy of one.

Aterio's October 21st flyover found unfinished work at IREN's Texas campus. Business Insider reported that Aterio estimates roughly 140 MW at Nebius's New Jersey project could slip into 2027.

Memory sync via MCP is live, but Notion's own October 19th release still listed skills via MCP as coming soon.

Yang's /fuck-cancer skill organizes records into a five-part brief for caregivers, with explicit limits against diagnosis and treatment choices.

Anthropic says the 96-copy phage vault, made in E. coli in under four months, gives biologists a simpler system for a particle unexplained since 1986.

Bloomberg reports some engineers' daily caps fell to $200 in October, with $100 planned and $25 possible for other teams, even as Apple pushes wider internal AI use.

UC Berkeley's Sky Computing Lab used coding agents to synthesize six model-and-Mac-specific serving engines, reporting up to 2.72x faster single-user decoding.

Built on Thinking Machines' Inkling-Small, the model is free for paid Devin users through November 30th while its weights stay behind a gated program.

The public beta pairs Firecracker microVMs with automatic suspension and billing for active CPU use and allocated memory, inside the same workspace as a customer's existing Render app.

Corning is investing $585M in North Carolina, while Relativity Networks announced a $40M hollow-core fiber order in August.

Goldie and Mirhoseini's chip-design lab released a 65B dense model it says scored 80.5 on TerminalBench 2.1 and cost $32M to train.

Enrich Labs launched Agent Conversion Optimization on November 4th, tracking Muse, Instinct and Dots daily from $199 a month, capped at 100 initial customers.

The Verge reported the OpenAI update on November 4th, but the rollout date and first-party Images 2.6 documentation remain unverified.

A federal judge discharged the show-cause order on October 28th after Cutera counsel said a corrected Legora draft existed but the wrong PDF was filed.

Co-founded by Gabriel Pimenta, Dharma-AI built its 2B extractor on its own OCR model and trained it on 43,384 field-level examples.

Solo founder Shawn Pana, previously on growth at Browser Use, is selling agents persistent cloud phones after open-sourcing the harness for personal devices.