GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior

(lesswrong.com)

2 points | by eatitraw 11 hours ago ago