Drive a browser with GPT-3
Here's a demo: https://twitter.com/natfriedman/status/1575631194032549888
Lots of ideas for improvement:
- Better prompt
- Prompt chaining
- Make a recorder to collect human feedback and do better few-shot
- Better DOM serialization
- Let the agent use multiple tabs and switch between them
Improvements welcome!
- setup virtual env
./setup.sh
- enter the virtual environment
source .venv/bin/activate
- setup
OPENAI_API_KEY
export OPENAI_API_KEY='<your_api_key>'
- the run
./natbot