Gosh. I've just asked #ChatGPT to help me label topics from a topic model based on 10 topic-defining words and 5 examples of high-scoring sentences per topic. Can't fit the whole conversation in a screenshot, so just showing the output. I'm mindblown. This is for me the first example of an interaction with ChatGPT where I feel like it's not just simplifying my work, but doing something I was struggling with myself #nlproc
It's not perfect, e.g., there are some repetitions, partly due to the model itself not being perfect, and to ChatGPT missing some nuances in the examples. Still, most of them are pretty accurate, and -- after a bit of back and forth -- hitting a pretty good level of specificity.
@rockberta this is interesting. I would be curious to see the prompt you used. One idea: to test this on other (open) large generative language models