logoalt Hacker News

aszenlast Thursday at 11:20 PM0 repliesview on HN

Agreed it probably contributes to the model improving for all agents but crucially it is verifiably better against their own agent. So they get a good feedback loop to improve both