The file
Sourced AI briefs in publish order — newest first. Citeable originals live on the homepage under From Signal Desk.
Mistral's first forecasting models are small, run on ordinary CPUs and carry a permissive license. The leaderboard wins are thin, but the practical package is the real story.
Anthropic says it will commit 5% of its shares to an Anthropic Public Benefit Fund, subject to stockholder approval, with board-discretion top-ups of 2% for each 50% rise in value above an undisclosed baseline. Uses could include direct distributions; governance is unpublished, and the fund is an LLC.
Anthropic has onboarded six people from nonprofit evaluator METR with company hardware, office space and internal-tool access similar to its own risk teams. METR will publish its findings without an Anthropic veto, though Anthropic can require redactions in set categories and decides what falls in them.
Meta allows adult users of its 18-plus Muse app to have explicit text conversations with the agent, while adult creators interviewed by WIRED say their Instagram accounts face constant restrictions. Meta says private chat on an adults-only app differs from marketing on a social platform.
NVIDIA's new Structured Data Models library makes table predictions without a training step, and a hands-on walkthrough shows the real risk lives in quiet defaults, not the models.
Meta removed a line in its consolidated commercial terms that said it could use content from a business account’s domains to develop and train its products. Legal experts had warned it could cover any publisher with a business account and override robots.txt. Meta says its crawling never changed.
NVIDIA says RTX Spark laptops are available from ASUS, Dell, HP, Lenovo, Microsoft and MSI, starting at $2,399, pairing a Blackwell RTX GPU and Grace CPU with up to 128GB of unified memory. Its launch recap covered six city events; performance claims are peak figures awaiting independent testing.
Kyutai's REKI shows a small robot model trained only on public data can remember what it just saw for almost no extra cost, and that memory stops arms from dithering.
Anthropic published a report on its workforce: junior staff in technical departments fell from 21% to 14% since early 2025, junior technical hires from 27% to 9%, and meeting time per employee rose 20%. A fifth of surveyed staff said their job shifted from doing work to managing Claude.
After saying onstage that Replit no longer works with Anthropic, CEO Amjad Masad clarified that users can still pick Claude, but OpenAI is now the default and preferred partner for Replit Agent. Since most users keep defaults, the switch could cut spend at what Masad called a former top-five Anthropic customer.
The Verge reports users of Meta’s Muse, xAI’s Grok, OpenAI’s Dots and Instinct have hit blocks on major sites. Yelp only allows agents that pay for data licensing, Delta blocks agent bookings, and JPMorgan Chase and Charles Schwab cite security. Partnerships are emerging as the workaround.
DeepSeek’s V4.1-Flash-1017 checkpoint posts 77.0 on DeepSWE and opens Search/Fetch APIs — a cheaper agent stack behind one key.