Anthropic and OpenAI want to embed safety evaluators
The frontier labs' push to embed independent safety evaluators is facing hard questions: will auditors paid by the companies they audit really be independent? The governance debate sharpens.
AI labs want in-house auditors
TechCrunch's second angle on the same story: before AI labs invite auditors in, maybe they should shut the front door first — securing frontier models against unauthorised access is the prerequisite nobody wants to fund.
Google Home
AI agents can now control Google Home devices — agentic AI moves from software setup into the physical smart home, with all the access-control questions that implies.
Claude chat and Cowork
Anthropic is merging Claude chat and Cowork into a single interface — another consolidation play in the race to own the AI workspace.
perv glasses
After accusations that its smart glasses enable covert recording, Meta is preparing to sell a camera-free pair — a rare public walk-back driven by privacy backlash.
Robots are waiting
Nvidia's Les Karpas, speaking at TechCrunch Disrupt, argues robots are still waiting for their ChatGPT moment — hardware capability isn't the bottleneck; the right model architecture is.
Also trending on Hacker News
- Training a 4B model to produce 81% faster query plans than Postgres — small models eating classic systems problems continues to be the sleeper trend of the year. Source
- Nvidia announces native GPU programming in Rust — Rust's systems-language moment arrives on the GPU. Worth watching for anyone whose compute stack is C++-shaped. Source
- Breaking the 1.58-bit barrier for ternary LLMs — quantisation research keeps pushing the efficiency floor down. Source