I don't wanna sound like a PM but it's literally the top part of results which can be removed with adblock plus element blocker.

It's an entirely optional addon added by google.

It's not just that. It's also embedded into many of the other results now. I see it in the middle of stock ticker results, local places, etc.

(I just deleted a duplicate comment to this). It's also in the research pane of docs, and a ton of other places.

And this is were the misconception appears. LLM's are a type of AI, but AI != LLM. The top part of the results, also called the AI summary. Is _not_ the only thing that is AI about search. Embeddings, a commonly used element of search, or ranking algorithms, use AI as well. More specifically, the techniques are known as machine learning, but it is AI. If you truly want to have no AI at all, then you are back to boolean retrieval systems. That basically boils down to: Count the amount of times the terms of your query appear in the document and return the top x results (which is impossible to do at Google's scale).

oh my god we're all talking about the AI summary results

Though note that before AIOs Google did also have LLM-powered text box results ("web answers") and nobody seemed to have a big problem with those... It seems a bit hysterical to demand them removed when they are just a better version of a long standing feature.

That was before the religious war.

They also sucked big time.

> That basically boils down to: Count the amount of times the terms of your query appear in the document and return the top x results (which is impossible to do at Google's scale).

It is very possible to do at google's scale. Jeff Dean and co built Map Reduce to help implement algorithms of this type. For example, https://en.wikipedia.org/wiki/Tf%E2%80%93idf is entirely based on counts. Combined with a sparse encoding and distributed lookups, this is how ALL retrieval worked for decades. In fact, any large scale retrieval system today uses a combination of sparse searches followed by a ranking or scoring with dense methods. You can roughly call sparse methods count-based and dense methods neural network based.

Source: the field of information retrieval https://web.stanford.edu/~jurafsky/slp3/

You’re right that there are other kinds of AI frequently used in search, but why do you claim having an inverted index is impossible at Google’s scale? How do you think exact term search works?

Are you suggesting that Google's switch promises to disable the use of all "AI" techniques down to the implementation level? That would imply it also disables spam filtering in gmail, which I haven't seen reported anywhere and isn't mentioned in the original Google post.

I don't think that's how they're thinking.

If search is that tightly integrated with the LLM capabilities isn’t that an architecture failure? LLM distends rely heavily on information retrieval for grounding their responses. Why would it be so difficult to let humans continue to directly use the information retrieval systems (classic Google) and keep the LLM systems optional and loosely coupled?