Recently Large Language Models (LLM's) jumped the threshold of being useful for problem solving in mathematics. It is expected that similar progress will happen also in theoretical physics, and the first signs of this are starting to appear. In our research group we wanted to find out what an LLM can do as a research assistant. We found that the current technology is at the level of a good PhD student or early postdoc, in problem solving, coding, and manuscript preparation. We obtained new scientific results that are worth publishing, regardless of whether or not they are calculated by the AI. In this talk I will share our experiences, and discuss the opportunities and problems that this tremendous progress brings about.