So the feedback loop on the large language models is making the models more bad. To satisfy the desires of the majority of people who use ChatGPT to figure out how to live, it's responses are getting worse over time because it's programmed to please the user.

