The relief-seeking behaviors were observed after steering and fine-tuning interventions, meaning the responses do not represent how public AI chatbots would behave.

Large language models (LLMs) are advanced AI systems trained on massive text datasets to understand, generate, and reason about human language, powering applications such as chatbots, code generation, summarization, and scientific discovery.