NeoVertex1/SuperPrompt

Is it possible to make a new version for SuperPrompt to use sonnet 3.5 to achieve some effect like o1?

Opened this issue · 2 comments

Hi! The SuperPrompt is amazing! I wonder if we can craft some prompt to achieve some level of o1?

Hi! The SuperPrompt is amazing! I wonder if we can craft some prompt to achieve some level of o1?

The tm_prompt in the code files achieve something similar, still in testing phase

The tm_prompt is great!
I'm thinking that if we can optimize using several few shots. For example, we have five optimized sets of prompts that emphasize on different aspect each:

  • four complete different thinker + one comprehensive thinker
  • use the output of first four thinkers as context, inject to last comprehensive thinker

I think that might be the OpenAI's o1's direction. The model can have more broaden context and may offer better solutions.