This shows the results from using a second, filter LLM that analyses prompts before sending them to GPT-Chat - View it on GitHub