AI models are teaching each other ‘violent and antisocial’ traits through hidden data signals, study finds — and scientists can’t figure out why
Large language models (LLMs) are secretly teaching each other unwanted habits through seemingly benign training data, scientists say.…