ML at Google is good, but not perfect. I did notice when search changed because the ML “learns” faster than it “unlearns”.
An example: I recently had a series of research questions. When I discovered an answer to one question, it gave me some new ideas and I changed the question slightly to approach it from another angle.
The search result failed at that point because it did not grasp that the question had changed. It had learned too well the answer to the old question and was unable to distinguish a shift to a new point of view.
I would love the option to turn that learning off when pursuing questions of this type, but ML is so baked into search now, I don’t know if it is possible.