To be fair, the single most publicly-accessed example of "AI" in the world can't find its ass with a flashlight and it's right on top of the search listings, where the most accurate search result used to be. Every LLM suffers from token drift over time so it's only a question of WHEN they're wrong.
To be fair, the system powering Google search is not the same one solving open Millennium problems. That’s like concluding that Concorde was a hoax based on riding in a Cessna one time. Not saying SI has been achieved, but you’re not even looking at the right entity to conclude one way or the other.