I think the bigger problem is how does something like ChatGPT fit into Google's business model for search.
Conversational interface and/or giving immediate answers may not be the best option in terms of ad revenue. Users re-iterating their search queries gives more opportunities to show ads. When users are scanning the page for best results, they might be more likely to click on the ads with clever wording.
> I think the bigger problem is how does something like ChatGPT fit into Google's business model for search.
Alternatively, how does ChatGPT fit into the usage model for search? I know it's taken on faith in this forum that somehow Google[1] search is bad or failing or about to be killed or whatever.
But... it's actually just fine? It does what the market wants, more or less perfectly. Everyone uses it to find the 99% of stuff that everyone searches for (commercial products, news, sites-you-forgot-the-name-of, etc...). What exactly is AI going to do better when all I want to know is where the Steam page for Skyrim is, or how much a pour-over coffee filter goes for?
The truth as I see it is that "search" simply isn't a market that's disruptable in the way posters here want it to be disrupted. Find 10 people on the street and ask them whether they can find stuff on the internet and they'll say "yes, duh?". This is like asking how AI is going to disrupt freight rail or high intensity agriculture. Those too are stable, already-optimized industries without a lot of room for genuine improvement except on cost (and recognize that training and running models of this scale is still significantly more expensive than crawling the web!).
Now, none of that is to say that there isn't an application for a generalized text model that can answer questions in the market somewhere. I think there probably is. I just don't see how it has anything to do with "search". Having a personal butler to do your research doesn't get you to the Skyrim forums any faster.
[1] Disclosure: I work there, but on firmware and nowhere near search. Opinions entirely my own.
> Everyone uses it to find the 99% of stuff that everyone searches for
You are probably right. What you are missing is that searches currently being made are a small subset of total searches that can actually be made (my guess is <10%), and will be unlocked by this new paradigm search, enabled by LLMs.
For example, Google is 25 years old, has the smartest people, best search technology and yet can not give an answer to a simple question my kid would ask like: "Do more people live in Madrid or Tel Aviv?"
What it can is give you ten links, then you open link #1, find out how many live in Madrid, then go back, open link #2, find out same for Tel Aviv, and then figure this out on your own. This is clearnly not the best way to do this, agreed? AI will do all of this automatically and unlock a whole new category of "searches" that did not exist before.
Another example is "Give me the names of all Texas-based CEOs that run companies with more than 200 employees.".
The only reseaon this search is not a thing today is because Google is not capable of satisfying it - at least not directly (20 searches and numerous ads later you might eventually find an answer, but that is not the point). Queries like this are a real need and belong in the vast pool of searches that will be (eventually) unlocked by AI.
Final example: "Summarize the latest research on quantum gravity for me". Does the user want to do 20 searches and read hundreds of pages of documents or they really want an AI to produce a summary with citiations, 5 seconds later?
Think of Google today as a best Nokia before iPhone showed up.
> if alternative business models haven't disrupted search in the last 20 years, why are they going to start now?
Because now is a good time, disruptive technology is here, and 10 links supported by ads and tracking had a good run, but the civilization is moving beyond.
> Alternatively, how does ChatGPT fit into the usage model for search?
I don’t think it does fit into it, but search is a means to an end. What people want is to find stuff.
For that, it might work. I also agree with you that Google already works for it in many cases, though. For example, if you ask Google for the weather in Foo, it first shows you the weather in Foo, then ads from advertisers who claim they can answer that question, and only then search results for hopefully relevant sites.
You don't think a ChatGPT product would... show ads? Something has to fund it, right? I mean, I get that the response will be that it will be supported by some other business model (hardware sales a-la Apple, windows licenses for MS, etc...). But... then it's not an argument about AI at all but about business models.
So... if alternative business models haven't disrupted search in the last 20 years, why are they going to start now?
I just think this is turning into an HN echo chamber. AI will absolutely be revolutionary and disruptive, but it's a long road from "revolutionary disruptive technology" to "Mighty Hammer That Will Destroy All My Imagined Enemies".
I agree with you, but Google is also a leading player in the conversational interface ("Ok Google")and in business software. Two areas where language models could have a huge impact.
Is Google deliberately slowing down any major improvements in these areas because they are petrified about the idea of losing revenue in their ad business?
Conversational interface and/or giving immediate answers may not be the best option in terms of ad revenue. Users re-iterating their search queries gives more opportunities to show ads. When users are scanning the page for best results, they might be more likely to click on the ads with clever wording.