Cue emphasized this shift as a technological inflection point, suggesting that AI could do more to disrupt Google’s monopoly than any court ruling. Alphabet (Google’s parent) stock dropped over 7% after Cue’s remarks, reflecting fears that AI could meaningfully weaken Google’s search ad dominance.
Cue’s comments are what one would expect to hear from a current senior VP of a company that has grossly underinvested in generative AI for the last few years because it failed to perceive the historic magnitude of the generative AI irevolution. For example, during 2024, Apple only invested about $4 billion in AI development and infrastructure, a year in which Amazon, Meta, and Microsoft/OpenAI invested seven to ten times that amount.
Now here’s a more plausible explanation of the decline in Safari’s search volume. In 2023, Perplexity introduced a new paradigm that challenged the old paradigm of Internet search, invented by and long dominated by Google.
Under the old search paradigm users provided a few keywords and Google returned a long list of links to websites that contained information about those keywords, with the best sources at the top of the list on the first page.
By contrast, under the new paradigm, a user asks a question and using a generative AI model, a chatbot returns an answer to the question plus links to reliable sources of data that support the answer. The new Q&A paradigm reflects how most people really think. They do not think about keywords; they think about questions for which they would like answers, Q&A.
Note that the sources of the answers do not have to be the best sources; they just have to be good enough to support the answers. So ChatGPT using Bing for search and Claude using a small company called Brave Search, can provide answers that might be good enough.
The irony is that Google has fully embraced the new Q&A paradigm, but it has two options.
Google search now offers AI overviews at the top of its cluttered first search results page on Safari.
And its Gemini chatbot can provide concise answers to questions in the same way that ChatGPT provides its answers because Gemini has an app just like ChatGPT’s.
However, given the special relationship that OpenAI enjoys with Apple at this time, most users of iPhones are more likely to be far more aware of the ChatGPT app than the Gemini app.
So we shouldn’t be surprised that Internet search on Safari declined. Savvy computer users everywhere are moving to the new Q&A paradigm. Indeed, Google’s latest quarterly report declared that its AI overviews had 1.5 billion monthly users, but obviously not on Safari.
Although the sources of a chatbot’s answers don’t have to be the best sources, everybody knows that Google’s search engine can find the best sources. Given a choice between good sources and the best sources, most savvy users will want answers that are supported by the best sources.
Therefore contrary to Mr. Cue’s dire predictions about the decline and fall of Google from its dominance of the old Internet search paradigm, Google might rise like a Phoenix to even greater dominance of the new Q&A paradigm. This prediction is persuasive, except for one unpredictable variable — the complexity of the questions as perceived by a chatbot. Consider the following example.
The editor of this blog recently noticed a couple of similar slogans used by the members of a political faction in this country with slogans used by members of a political faction in another country. So he asked ChatGPT running on GPT-4o, which is not supposed to be a thinking model, to determine whether there were any formal connections between these two factions.
Had ChatGPT been able to search The New York Times, the unofficial national newspaper of record in this country, this might have been an easy question to answer.
If ChatGPT said there was a formal connection, it would have produced links to articles in the New York Times about the formal connection. If it said there was no connection, the editor would have been inclined to accept that judgment, even if, in fact, there was a formal connection the New York Times reporters had not discovered.
But ChatGPT cannot search the New York Times archives because of the lawsuit that’s still going on between OpenAI and the New York Times.The editor was surprised when ChatGPT suddenly went into chain-of-thinking mode, examining lots of sources that he regarded as not the best, certainly not comparable to the New York Times. He tracked its chain of reasoning for 15 minutes, after which it suddenly produced a long, but persuasive summary of its findings.
ChatGPT found that there was no formal connection between the two factions; the connection was ideological, a conclusion that was supported by ChatGPT‘s citation of many slogans, phrases, and hats used by both factions and by explicit references by the faction in the other country to the faction in this country.
This example shows that a chatbot with weaker sources might be able to compete with Google’s stronger sources by applying more complex thinking to analyze its weaker sources. And that’s why this game is not yet over and it won’t be over until it’s really over, no matter what an ill-informed Apple senior VP believes.
Cue’s misinterpretation of the decline in Safari’s search volume is a glaring example of how little Apple's current leadership understands about the generative AI revolution. But perhaps the best example is the simplest: their characterization of Apple Intelligence as an “upgrade to Siri”.
No comments:
Post a Comment
Your comments will be greatly appreciated ... Or just click the "Like" button above the comments section if you enjoyed this blog note.