Ask HN: Looking for recommendations on clustering similar LLM questions

Looking for recommendations on clustering similar LLM questions for trend analysis. I'm considering two approaches:

1. Vector space embedding + n clusters -> sample 10 questions per cluster for LLM summarization. Works well when themes/intents aren't predefined.

2. Using LLM to predict question intent/labels. More expensive but better when we have defined themes.

Which approach would you recommend?