← Monday, August 31, 2026
AI Frontier · Aug 28
ModerateAnthropic shows AI that can improve its own training
Anthropic ran a test where AI helped make AI better. The system read past research, tried new ideas, and checked the results. It did this on its own, like a small research team. On 10 safety tests, the AI beat what skilled humans came up with. It did it in about six hours, for far less money. Anthropic says the model did not get worse as it improved.
Why it matters to GCs
AI is starting to speed up its own growth, so tools may get better fast. Better tools could reach the job site sooner than you expect. It is worth watching so you are not caught off guard.
TechCrunch·Via POLR AI Hub