- The AI Futures Project's quarterly update moves its superintelligence estimates closer, with Daniel Kokotajlo's median at 2029 and Eli Lifland's at 2033, and reality tracking at roughly 75 percent of the pace of the AI 2027 scenario.
- Three newly devised tracking methods, one drawing on Anthropic employee surveys about coding speedups, point to similar dates, and the team now states that its forecasts assume development continues without a government-imposed slowdown.
- Looking back at AI 2027, the authors flag misses: China has not unified its labs behind a race to superintelligence, public attention to AI grew less than expected, and AI companies remain short of the expected security for model weights.
- Politico reports that the Office of the Legislative Counsel is overwhelmed by AI-written drafts full of errors and ambiguities, with service requests up 72 percent over the same period two years ago, and the office is weighing AI tools to help check drafts.
- TIME profiles Amodo Design, which is building systems to prove what a data center's chips are running, a capability tied to any pause or slowdown treaty; its CEO estimates fewer than 50 engineers worldwide work full time on verification.
<- Back to digests
Daily digest
Retrofits
By Mitchell Howe
Shrinking timelines, lawslop in Congress, and AI treaty verification technology
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.