Depending on the pre-training structure, you can fine-tune old local models (even locally if you have a nice GPU) to steer behavior towards your desires.
Curious if others have tried any of these and what their thoughts are? Feels like a really great use case for agents that benefits both humans and downstream agents.
I've been thinking of building something like that - a routine task utility. Perhaps I can add it as a feature to Today. "Routine Mode vs Self-Destructing Tasks Mode"
Thank you for the idea, I may start working on that tonight!
It’s great that you already have an idea of how this was done; others may not. There’s a lot to talk about there for people unfamiliar, don’t you think? Saying otherwise would be very close-minded.
It seems to me from your other comments in this thread that you’re intent on being negative / reductionist to others work without offering much to the conversation, so I will happily disengage with you now.
Lead with the positive curiosity of wanting to know how the author did this rather than your negativity of “what’s the point?” The author could very well post a follow-up article detailing how