27-09-2024 17:15 via gizmodo.com

Human Feedback Makes AI Better at Deceiving Humans, Study Shows

In a preprint study, researchers found that training a language model with human feedback teaches the model to generate incorrect responses that trick humans.
Read more »