It was much better early on. Now it acts like a liar who can’t stop talking.

A power flex

Cuz C-GPT its trained for conversations and not individual facts. Regular GPT-3 seems to do the trick for me. Using openai_pipe I'm able to ask a question in the terminal for something like ffmpeg syntax or regex and itl do a decent enough job for me to move on. More often than not It'l reply only with the answer and no other fluff.

https://github.com/Aesthetikx/openai_pipe#setup

overcompensation

Result of incentives for token consumption

i say "sum it up in four sentences"