3 Comments
User's avatar
John Damask's avatar

Terrific write-up. Thanks for sharing.

At the risk of infinite recursive time-wasting, I wonder if GEPA could be used to tune feedback in the optimizer metric

Expand full comment
Michael Pavlukhin's avatar

Probably we can do this. I had idea that we can train Judge LLM like we train GANs

Expand full comment
Martha's avatar

that’s very useful, thanks! 👍🏻

Expand full comment