Impromptu: Amplifying Our Humanity Through AI In my opinion, it’s a plus that LLMs have this imaginative/ conceptual capacity. Indeed, this is how I’ve spent a lot of time interacting with GPT-4 to date. Part of what makes it so com- pelling is that it’s just as ready to draft a new Star Trek: The Next Generation plot as it is to summarize a legal document. (See the chapter “Public Intellectuals” for more.) In the end, I assume what most people would choose is an LLM that functions in a reliably factual way in some contexts, in a more imaginative way in others, and is clear about which of these modes it is currently in. (Which, of course, is how we’d like the other humans we engage with to act as well, but which they don’t always do.) So for the time being, yes, we have to treat GPT-4 with the sort of caution we’d use with any adolescent technology (or for that matter, any adolescent human). But given how fast OpenAI is improving its models, I expect we’ll soon see a version of GPT that significantly decreases unintended hallucinations, while also preserving its ability to generate imaginative and conceptual texts. That’s where I have my own sights set for GPT: a tool that ven- tures beyond “good enough knowledge” and arrives at an even more impressive destination—a tool that functions both as a source of truth and a source of inspiration; that can not only fact-check our assumptions, but also spur us to think of new possibilities. 164