MarijnVeenstra
commented
1 year ago Researched this issue, found the following that can be used: https://openai.com/blog/function-calling-and-other-api-updates?ref=upstract.com https://www.mlq.ai/gpt-function-calling-getting-started/
It allows users to describe functions and have gpt4 output a JSON object containing arguments to call those functions
PIM can be quite good at holding a conversation and helping to translate sentences. The main problem is that it can be tricky to switch modes.
At the moment, PIM looks for the keywords 'conversation' and 'translation' in the user's speech to change the current mode. Although it may be possible (and if it is, it will be faster) to do everything with a single prompt that returns either an answer to the user's question, a translation, or a mode switching parameter.
It is important to understand that the workflow should also change - we get input from the user and transcribe it. With the input we should check if we need to change the mode. If it is the case - we return the mode change parameter. If it is not the case - we continue with the corresponding mode provided by the user (either "translation" or "conversation").