flat hatch
Hello and happy hackathon to everyone,
As part of developing a plugin for G-Assist, which can be divided into three main parts:
- retrieving data from external sources (within the plugin)
- retrieving custom data from the user's computer (within the plugin)
- analyzing this data and making recommendations based on the previous sources — what I will call the "expert logic."
I want the "expert logic" to be performed by G-Assist, specifically by LLaMA 3. To pass this specific expert logic, I need to provide textual information to LLaMA 3 that describes the rules for interpreting the data and the resulting recommendations.
My challenge is to find an effective way to pass this expert logic to G-Assist (LLaMA 3). This logic is a long text containing multiple rules.
The ChatGPT plugin "Project G-Assist Plugin Builder" helps a lot, but it sometimes contradicts itself (as often happens with AI) and offers me two approaches:
- pass the expert logic as text by adding it to the user prompt
- pass the expert logic in the plugin's response back to G-Assist along with other parameters (data)
I have tried multiple approaches, including prompts only (containing all data plus expert logic). However, I often encounter a problem where G-Assist frequently responds that it is designed to optimize GPU/system performance, or gives messages indicating it cannot answer my question, or sometimes returns answers that do use the expert logic but with interpretation errors.
My question is: do you have any recommendations for passing the "expert logic," whether as a long explanatory text or other formats if recommended, to G-Assist (LLaMA 3)?
I hope I was clear, and thanks in advance to anyone who can help with this topic.
Have a great day everyone!
tiny fog
