- A new watchdog group, Guidelight, graded five frontier labs on whether they can control the AI systems they build; no company scored above 3 out of 5 on any measure, with Anthropic and OpenAI tied at C+ and Meta last at F. (Donald)
- The review found detection and outside assessment are the strongest areas, while automatic shutdown mechanisms are incomplete where they exist at all, and Anthropic has no publicly known plan for handling a model that goes rogue. (Donald)
- Mathematicians at a summit questioned whether human expertise in their field will survive, an example of what the piece calls jagged intelligence, where AI improves at very different rates across tasks. (Joe)
- Medicine, hiring and physical work are also being encroached on, from a JAMA article on AI rivaling doctors at cognitive tasks to Chinese robot demos, while proposals such as AI dividends or basic income circulate for a future with fewer jobs. (Joe)
- Datacenter opposition is becoming an election issue in Pennsylvania and Wisconsin, and industry publicity campaigns, community funds and pledges have done little to shift opinion, with young Americans increasingly more concerned than excited about AI. (Joe)
<- Back to digests
Daily digest
Uncontained
By Donald Gauvreau and Joe Rogero
Bad grades, jagged job threats, and reputational damage control
Read the full dispatch on AI StopWatch
From AI StopWatch, published with their permission.
The summaries and the full Spanish translation are produced automatically. The original is always linked.