[throwitaway222]

I bet that the LLM can write a program that CALLS GPT and tells it to lie 20% of the time in the prompt

layers people