I agree, that language could be very improved. This is not a GPT-like LLM whose training corpus is HN comments, which I found to be an extremely interesting idea. Instead, it looks like it's finds relevant HN threads and tells GPT-3 (the existing model) to summarize them.
To be clear, I think this is still very cool, just misleading.