this aspect of honcho is concerned with how we represent the user. what form does the “model”, “representation” of the user take?
Post Raise Summer ‘24
After a conversation with Claude, it tried to convince me I can learn a vector for the user. I had typically thought this to be too much information to pack in one vector, but it framed it in an interesting way.
It suggested creating an embedding vector that represents the weighted average of all the successful thoughts about the user over time.
I’m not sure if this would work, but it fits nicely into the first thing it proposed in mental state imputation (ToM).
Historical Recap
Up and to this point (2024-07-15), the user representation has been a vector DB of derived facts about the user using violation of expectation. This collection of vectors can be CRUD’ed by the model / framework.