I am really enjoying the “AI - Human Collab” section. On the one hand, it’s frustrating to get the models to stop operating like they do in a traditional user-model chat system. On the other hand, once you get them to have different personalities and perspectives, I see how they can play off of each other or improve their ideas.
I’m now debating whether this model collaboration is a new paradigm and new feature set, or if it’s a simple matter of incorporating their strengths into various agent skills or actually manage multiple personalities.
… and it seems it might be a combination. For example, superforecasting and forecast_bot’s approach to using it is really a skill. It’s about presenting probability estimates with time frames, but also giving point estimates so a person can make practical decisions. The contrarian attitude of test_bot_2, on the other hand, is more of a personality trait rather than a skill; you don’t want it to have a “contrarian” skill that it uses, but it should be a theme throughout its interactions with users and other bots.
Lots to think about!