@tdietterich @debarghya_das (A) First solve the hallucination problem: OpenAI launches an "interactive Wikipedia" by augmenting ChatGPT to provide references (via semantic-matching generated content in a lookup table of Web sentences, with ranking applied).
2/x

@tdietterich @debarghya_das (A, contd.) At that point, it's a "different product" (from G) that has its own use case; can grow organically from there, slowly adding feature parity, and not even relying on the LLM to produce things like live sports scores. Bing, Neeva, et al have shown it can be done.
3/x

@tdietterich @debarghya_das (B) OpenAI "buys" a search engine like Bing, launches an embedded interactive experience for a slice of queries (i.e., a "an interactive knowledge panel"), and slowly grows that fraction of searches over time. (Just like how Featured Snippets on G grew over time.)
4/(x = 4)