This is a project I did that uses the OpenAI API to run debates betweeb two AI models. Instead of use the context window however, we used a form of homemade RAG. The idea is for previous debates to be used to improve on the current answer without resorting to context. Instead, we use a variety of metrics to determin the quality of the argument which is fed back to the model.
The "main" page is the page showing the current metrics and a few graphgs that were used in the report
50 Debates have already been run and are including in the existing database file. To run more debates, follow the instructions below.
- Add OpenAI API to key to .env
- Click the Debate screen and enter a question
- Specifiy Number of Runs to Make in .env file (include those that must already be run) if you want to run automatically, currently there are 50 debates in the database
- Click the 'Auto Run'