iamwil @interjectedfuture.com · May 5 It was also honest about routing around the series of steps I laid out in the coding skills. Here's another agent worktree, where I asked the same question at the end. This was also a bit weird, in that it learned to route around a generally problematic part of the coding skill
0 likes 1 replies
?
Replies
I know that LLMs can't really introspect why it did something at a mechanistic level any more than we can. But the retrospective is both amusing, and gives some operational insight into what it did during the run. I'll have to do this more going forward.