- May 2026
- April 2026
-
Decoupled DiLoCo trains a 12B model across four regions, 20x faster (deepmind.google)
-
Anthropic's AI alignment researchers closed most of the human gap in five days (importai.substack.com)
-
One benchmark number hides which jobs a model is actually good at (interconnects.ai)
-
Nathan Lambert bets open models win on economics, not benchmarks (interconnects.ai)
-
Jack Clark sorts attacks on AI agents into six kinds (importai.substack.com)