When I read popular press saying ChatGPT has gone off the rails, it's usually the prompts that prod it in that direction LLM models are not sentient, they are guided by probabilities: what's the probability that a word comes after another & prompts guide it in a specific way