Arjun Karanam Puts Expert Corrections Into the Training Loop
Trajectory's co-founder wants deployed AI agents to learn from expert edits. Its new graders and one-rollout research expose a harder question: who decides which corrections should change a model?
Read the article