vnoelifant
commented
9 months ago I am using a tool as part of a Marvin AI Application, called get_feedback_inquiry, which generates a feedback inquiry based on the user's request type and sentiment, but in my app, I call this function, and the function actually has Python logic. I am wondering if I am doing too much here, and if all this really should be part of some prompt engineering docstring I need to add.
See line 293, 76 in https://github.com/vnoelifant/cozmo-companion/blob/main/src/cozmo_companion/assistant.py.
Can I simply just describe to the AI model what I want to do here, instead of writing all this Python logic?
Example AI Application with a tool (from Marvin version 2)
It seems like in the example above, the tool is simply described in the GAME_INSTRUCTIONS docstring. Do I need to do something like this with my app? Maybe I don't even need a "tool" and I can just describe the expected dialogue behavior in the instructions
Per Marvin v2.1.4 update, https://www.askmarvin.ai/docs/interactive/applications/, need to refine application by adding instructions with state management to allow application to understand what to do based on docstring prompt engineering.