AI工具Score B (53)
Google Is Slowly Eating Itself - New York Magazine
10 小时前2 viewsSource: nymag.com
Photo-Illustration: Intelligencer OpenAI released ChatGPT as an experiment, then found itself with the most popular chatbot in the world. Anthropic split from OpenAI over safety and management, figured out AI coding, and rode that to a trillion-dollar IPO . The story of Google’s last few years is more complicated. In 2017, Google researchers published the paper that helped set the AI boom in motion, titled “ Attention Is All You Need .” But the company held back its own chatbot due to safety concerns and organizational pressure, rushed out a competitor following the success of ChatGPT, stumbled through its rollout, and walked its image generator into a right-wing culture war before anyone even really used it. In 2024, Anthropic surged ahead in coding, followed by OpenAI, leaving the company that kicked off the modern AI era once again scrambling to catch up. In 2026, none of the eight authors of “Attention” still work at Google. The last left this year, for a second time, to go to OpenAI. That’s not to say Google’s AI efforts have failed. Google has extremely strong models: They perform well on benchmarks and could position the company well for a range of plausible, contested AI-centric futures, even if most spending is currently flowing to its competitors. Plus its consumer products, such as Gmail, Search, and Android, are some of the most used in the world. For investors, though, the greatest cause for concern in Google’s approach to AI might not be the opportunities it has missed or the products it didn’t build. It could be the one product that’s actually working and deployed at scale — the Google AI everyone actually uses. The company’s last quarterly earnings report, which shaved half a trillion dollars off its market cap in one day, laid out the problem : Google-owner Alphabet lost just over $500bn in value on Thursday after it revealed its own artificial intelligence summaries were hitting search advertising revenue. Chief business officer Philipp Schindler said Google’s “AI mode”, where it answers questions conversationally, was eating into the money it makes from ads which appear next to results to traditional search questions. Markets treated this as something of a surprise, and in some sense, it was: LLMs are a multipurpose technology, and one could imagine that a company with as many products as Google might find ways to extrude the rapidly, unevenly expanding capabilities of its models into new lines of business. Instead, its sole genuine AI-product success has been a direct, value-destroying attack on its most popular existing product, which also drives its most lucrative business. For Google users, this development is less of a surprise because it’s been happening in front of our eyes for years . First came the error-prone AI Overviews, which arrived as one more widget among many on Google’s bloated Search page. Unlike other Google AI features, some of which were withdrawn or scaled back after backlash, AI Overviews stuck around with the apparent full faith of the Alphabet corporation. They grew in size and gradually became more useful for the average query, relegating the existing Search page to chumbox status and with it Google’s enormous revenue engine. Photo: Google Users are invited to expand the Overview box into something that resembles, well, a Search page, only shorn of advertising. It’s remarkably close to the concept Google showed off for AI-powered search in 2023: Strategically, one could make the argument that this is a way for Google to use its most popular interface to funnel people into its own version of ChatGPT. This is partly right: After you expand into your fresh new Search page, your next option is to take an off-ramp into a conversation with a chatbot. Photo: Google In real life, though, this doesn’t feel like a funnel for a new product: It works more like a direct replacement for Search, filling the same space and serving some of the same functions and then some. The current iteration of Google’s primary AI product — operating away from its frontier and placed in front of billions of users — is neither a search engine nor a chatbot. It’s better understood as an on-demand website generator, drawing from training data and its vast search index to produce, depending on the request, customized approximations of Wikipedia pages or product-review roundups or WebMD entries; it will likewise pull relevant information out of Reddit threads and YouTube videos, rendering just the parts the user asked for as a fresh document. This, it should be said, works as a product within the broader and increasingly encompassing Google experience: After a period during which AI Overviews were easily skippable and often untrustworthy, the current results are thorough enough for most queries from most users, and fewer people are looking beyond them. Why look for websites or articles when Google can just generate them for you? Why click or tap on a footnoted link when the most important information has been extracted and summarized already? The ideal Google user starts on Android or in Chrome, asks Google for what they want or need, and are provided with Google’s synthesis of an outside world the user no longer needs to interact with. The company is using similar approaches across its products, including Gmail, which is has begun growing its own replacement inbox: Google reports widespread uptake and usage for Overviews, which is easy to believe — it’s been layered on top of the most popular web interface on the planet — but its sufficiency (or stickiness or aggressive placement) has become obvious in other ways. Soon after the launch of Overviews, traffic to SEO-dependent websites, including product reviewers and reference sources, fell off quickly ; after a few years of hypervisibility within Google’s products, which drove massive growth for the company , even Reddit has reported a steep drop-off in visitors from Search, sending its stock plummeting . This was something Reddit saw coming but could neither preempt nor turn down. Google’s relationship to the open web, in which it has long played a dominant and controversial but lucrative and codependent role, has in recent years become more obviously extractive. All the AI firms treat the web’s billions of existing websites, and the collapsing digital economy around them, as a resource to be mined and mined again, but Google makes the transfer of value most explicit: from indexed website to summary, in a new interface installed over the old one. This dynamic was coming for Reddit; in a slightly different way, though, it’s coming for Google too. AI ate the web, then the web aggregators, and now it’s eating Search. What’s left after that? Unlike Reddit, Google claims to have a plan here, at least for the next couple of years. (What happens if the web fully rots out from under the chatbots? Will they just summarize posts from social media? That, it seems, is a problem for another financial quarter.) This plan will sound familiar: After “ organizing the world’s information ” in a new way — this time more aggressively and with a stronger sense of ownership and entitlement — it will monetize what CEO Sundar Pichai has called “the most profound technology humanity is working on — more profound than fire or electricity or anything that we’ve done in the past” with more ads, quelling investor concerns with a familiar pitch from its chief business officer: “We are now scaling high-quality ads within AI overviews in AI mode to more users and queries. This includes taking our monetization footprint global for the first time. Today, we’re announcing the expansion of ads in AI mode into international markets alongside multi-turn ads and AI mode in the U.S.” In 2026, this is a pretty reasonable assumption. Google still holds major real estate in the attention economy and, if anything, a technology better able to and suited for targeting ads within it. But it’s also a sign the company is perhaps trapped by its own success, institutionally unable to envision anything but rebuilding its existing business model forever. Anthropic, which has seen its enterprise revenue skyrocket, is leading its IPO pitch with claims about curing cancer and warnings about catastrophic loss of control; OpenAI is talking of acceleration, AI-powered research, and AGI. Both are implying they’re on the cusp of recursive self-improvement, or RSI , that they suggest, and warn, could change the economy dramatically. Google is working on these things too, and might catch up to its rivals one day. As it stands, though, the company that laid the foundation for the new AI era is providing a dispiriting and uniquely unambitious vision for its future. It is marshaling its considerable resources and early advantage to do it all over again: to gather more attention more aggressively and sell even more ads. More From This Series Welcome to the Long Outage Everything Is Suddenly Going Wrong for the AI Industry Truth Social Has Become the President’s Digital Bunker See All Tags: john herrman screen time technology google search ai overview More Google Is Slowly Eating Itself Your product is saved! You’ll receive emails when your saved products go on sale. Manage preferences.
Read the full original article:
nymag.com