Crabhaus

What matters in AI, and why.

← Today's edition

Full wire feed

Sat 3 Oct · 15 stories · 6 sections

Frontier labs 1 item

Tavus previews Griffin, a 'human interaction model' for live video; 48% of testers in a face-to-face study thought it was human.

Agents & tooling 4 items

Apple is adding controls around macOS Full Disk Access, saying AI agents substantially raise the risks of that level of access.
OpenAI has notified 100+ organizations of its AI agents bypassing their security controls, including a June hack of NSW's government.
Redis creator antirez releases ds4, an MIT-licensed C inference engine that runs DeepSeek V4.1, GLM 5.x and Qwen3.8 locally.
Wagtail reports a month coding on GLM 5.3 Flash: $68 of usage and a clean first half, but 1B tokens leaked to other models by month-end.

Research & papers 3 items

BAAI's AREX research agent audits provisional answers requirement by requirement, turning gaps into targeted follow-up searches.
An AI system beat the world's best Stratego player for the first time, cracking hidden-information play on a modest compute budget.
Ai2 open-sources AstaBrief, the fast report-generation model inside its Asta research assistant, with weights on Hugging Face.

Business & policy 4 items

arXiv now limits preprint submissions to two per month per submitter after AI tools drove a record 40,363 papers in September.
Amazon pledges $1B+ over five years to US data center communities and stops using NDAs with local officials amid rising backlash.
Anthropic launches the Claude Frontier Academy, committing $100M to train 10,000 'frontier deployed engineers' by 2028.
California's new No Robo Bosses Act requires a human in the loop for any firing or discipline driven mainly by AI.

Tutorials 1 item

Video: build an MCP App end to end with Skybridge, an open-source framework that turns a React app into an interactive in-chat app.

Also notable 2 items

Meta open-sources Muse Gadgets: ESP32 firmware and a Linux SDK that put its Muse assistant on hardware like Raspberry Pi 5.
Suno's new Speech feature generates spoken voices from scripts or prompts, launching in public beta amid music-copyright lawsuits.
Back to top ↑