Who read it? LLM may be great at coughing up seemingly good long chunk of text, but human is not capable of reading all of these verbose text because of the simple fact that GPT can generate texts faster than human reading speed.
So I bet what will happen in that case is, humans use LLM to summarize these LLM-generated verbose documentation. What a wonderful bureaucratic world will that be?
Tons of pollution for this shit.
If you go through my history here on HN (you don’t have to I’ll sum it up), you’ll see that I’m not a fan of LLMs for code generations. In fact I’m not convinced they will ever be good enough to be useful for it. For documentation, however, at least for JSDoc GPT is incredible. I really mean it when I say that it writes it better than we do. It’s able to figure things out and describe it in ways that still amazes me from time to time.
It does require you to be good at naming things.