In general I agree with you, but sometimes the “live”ness of demos is essential. I’ve given quite a few presentations about LLMs over the past year and a half, often to audiences unfamiliar with or doubtful about them. Being able to have on-the-spot, unplanned interactions with an LLM has been very useful for showing listeners that they are not just preprogrammed algorithms.

But I’m giving another demonstration in a couple of weeks, and this time I will be ready to show not only ChatGPT but also Claude and Gemini. And I’ll have some previously prepared examples ready in case the Internet connection goes down.