Perform facts checks on your conversations with LLMs to catch fake-news, misleading information, and LLMs confusion. FactsCheckerGPT enables you to talk to ChatGPT, and to validate each response against an "army" of LLMs. As the FactChecker Agent2, the user can choose a different LLM, such as GPT3.5 / Alpaca (more to be added soon). Currently, there is no support for multiple FactCheckers.
Run the facts_checker_gpt Python module in your terminal:
(Type this into your CMD window)
python -m facts_checker_gpt
You can start asking questions:
To use Alpaca, follow the getting started guide of alpaca.cpp.
After you've downloaded both files, update your .env file:
ALPACA_MODEL_PATH=C:\alpaca.cpp\chat\ggml-alpaca-7b-q4.bin
ALPACA_EXECUTABLE_PATH=C:\alpaca.cpp\chat\chat.exe
AGENT1_MODEL=alpaca
AGENT2_MODEL=gpt-3.5-turbo-0301Currently, the alpaca model doesn't perform well as a verifier (tested with the version of the 7B parameters).