Are we sure this deal is about answering new SO questions with LLM? It's more likely to be a deal where SO sells access to its database to OpenAI so they can use human-generated content for LLM training, and SO gets to use LLM as a more efficient search through its human-generated content.

It's possible they could also choose to delegate the duplicate decision to the LLM but let's be honest, that decision is currently crap anyway.