@abacaj - Here's a snippet of using gpt-4 as a "reward" model. Has

Here's a snippet of using gpt-4 as a "reward" model. Has been working *really* well for me (better than using a numbering system)