Not just in your head: ChatGPT’s behavior is changing, say AI researchers

The paper notes the opaque view of closed LLMs like ChatGPT, and says improving monitoring and transparency can avoid pitfalls of LLM drift.

ai ai researchers arxiv automation behavior berkeley berkeley ai research big-data-and-analytics business chatgpt conversational ai data infrastructure drift gpt-3.5 gpt-4 head llm llms ml and deep learning monitoring nlp notes openai paper programming & development researchers stanford university transparency