Time on task only matters for the lowest paid. At high levels robust, predictable, and other characteristics compete strongly for value. Do you want the most accurate medical diagnosis or the fastest?
So you want the answer to when the taco stand opens to be quick and wrong?
That's possibly the biggest waste of my time you could think of, because I'm probably going to have to spend a half-hour and a trip out of the house finding out that the answer was wrong. I'd rather get the answer in 5 minutes and for it to always be right. Dying in an accident coming back from a taco stand AI didn't know closed 6 months ago would be the worst death.
There is no "typical" search engine use case. The fallacy that there is such a thing is a big part of the problem. The enshittification and decline of the digital window to the internet is so complete that even basic information management tools like browser bookmarks are deprecated: people will "search" even for sites they use repeatedly.
Minimally there needs to be a transparent split between commercial queries (searching to buy something) and knowledge / abstract queries (searching to learn something).
Users should be context aware (ideally using completely separate tools) of when they are simply accessing a pay-to-play online product catalog versus when they are querying what is effectively a decentralized wikipedia.
Commingling the commercial with the factual was always going to be a dead-end.
you get wrong answers from either source. When ai builds in fact checking that's accurate to validate it's answers then it's game over and that wouldn't be too hard to implement and future versions will have a lot less hallucinations and faulty facts.
