Observation Model

[matfat55]

This is a feature that kodu.ai's extension has, here's the src for it

A secondary API request every x requests to another model which rates and scores the current model's performance in the last y messages (x and y can be set to not be the same). It suggests improvements and helps guide the main model. The feedback and score is shared with the main model.

[hannesrudolph]

@cte what are your thoughts on this?

[cte]

Pretty interesting, and I think related to sub tasks and custom modes. It seems like you can approximate this behavior with a mode that spawns tasks and rates the performance.

[BradKML]

vibes like Task Master and the way they handle complexity, but they are not made to self-optimize and have critical thoughts. Those are more delegated to some implementation of memory banks