AI has spread into search results, supermarkets, movie theaters and home robots — and Google is paying billions of dollars in lost ad revenue to keep its AI Overviews in front of users, MIT Technology Review reported on Tuesday, November 18, in the latest edition of its AI Hype Index.
That figure is the sharpest number in the report, and it cuts against the industry’s own story. Google is not deploying AI Overviews because they print money. It is deploying them because it believes the alternative — a search page that looks like 2015 — is worse.
When the company distributing the technology eats a multibillion-dollar cost to force adoption, that tells you the deployment is defensive, not organic. MIT Technology Review frames the whole moment with a line that lands: like a nasty rash, AI seems to spread everywhere.
Google pays for its own disruption
The economics deserve unpacking. AI Overviews sit above traditional search results, which means fewer clicks on the ads beneath them. Google has accepted that trade anyway, according to the report, because the answer-engine paradigm — where a model synthesizes the web instead of linking to it — is now table stakes. The company that defined the link economy is dismantling it from the inside, and billing the loss as a strategic expense.
That is not a hype cycle. That is a repricing of the entire search business.
The report fills in the physical world around that digital one. Two people hauling an AI-powered humanoid robot up a stairwell to clean someone’s apartment. A supermarket floor slick with melted ice cream, the aftermath of what the report calls an AI-assisted cyberattack. Neither anecdote is glamorous. Both are the point: the technology has stopped being a product category and become ambient infrastructure, arriving whether or not it works well.
Ten years after Lee Sedol
The most technically loaded claim in the report is also its quietest. Ten years after DeepMind’s AlphaGo defeated Go champion Lee Sedol, MIT Technology Review argues, today’s large language models still aren’t tapping into the machinery that made that win possible.
This is a real architectural distinction, not a rhetorical one. AlphaGo paired deep neural networks with Monte Carlo tree search — a system that could plan, simulate and verify moves against a defined ruleset before committing. LLMs generate tokens sequentially with no ground-truth verifier attached.
The report’s companion piece, titled “Don’t be fooled — LLMs don’t reason,” drives that home: fluency is not planning. A model that produces a correct chain of steps is not the same as a system that searched for it.
Why does the distinction matter now, a decade later? Because the industry’s current bet — test-time compute, chain-of-thought, “reasoning” models — is essentially an attempt to retrofit search-like verification onto architecture that lacks it.
The report’s skepticism suggests that retrofit is hitting diminishing returns, and that the breathless AGI claims of the past summer “fall apart pretty quickly under scrutiny,” as MIT Technology Review puts it in its own framing of the summer’s hype.
A 26,000-word post upends the argument
The counterweight to that skepticism is the report’s exclusive interview with Nick Cammarata, the researcher whose 26,000-word viral blog post has, per MIT Technology Review, AI safety people celebrating. The post’s headline claim — “We were looking at it wrong” — is framed by the outlet as an explosive claim that has upended the AI world.
The tension here is structural and worth sitting with. The same publication, in the same index, is simultaneously publishing that LLMs don’t reason and that a viral reinterpretation has safety researchers celebrating. Those two pieces are in productive conflict.
One says the machinery isn’t what we think it is; the other says our way of looking at it was wrong. The AI Hype Index exists precisely to hold both without resolving them into a press release.
Grok's Odyssey feels like an unskippable mobile ad
The culture-war layer gets its own entries. Andrew Garfield plays Sam Altman in a film the report describes with the word “nauseating” attached to his intensity. Elon Musk has released a Grok-fied version of The Odyssey — and the report’s verdict on the viewing experience is blunt: it feels like a long, unskippable mobile ad.
That last phrase is doing real analytical work. When a generative-AI artifact is indistinguishable in texture from an advertisement, the technology has failed at the one thing its proponents promise — that it unlocks new forms of expression. A 2,800-year-old epic run through a model produced something with the pacing of an interstitial. The tooling is everywhere; the output still feels like marketing.
The agentic web, and who it’s for
The report’s forward-looking entry is its most consequential for anyone who builds or operates online: the agentic web is here. With the rise of personal AI assistants, MIT Technology Review states plainly, the internet isn’t just for humans anymore.
Read that against the Google figure at the top and the dependency chain snaps into focus. If traffic shifts from human browsers to autonomous agents, the ad-revenue model that funds most of the open web loses its audience.
AI Overviews are Google’s hedge against that future — a way to keep the human in the loop on Google’s own page. The humanoid robots in the stairwell and the agents crawling the web are the same bet at different scales: replace the human interface, and charge for whatever replaces it.
- Search: AI Overviews persist despite costing Google billions in ad revenue, per the report.
- Architecture: a decade after AlphaGo, LLMs still lack the search-and-verify machinery behind that win.
- Culture: an AI-made Odyssey and an Altman biopic mark AI’s arrival as mainstream entertainment — with mixed results.
- Infrastructure: personal AI assistants mean the web is no longer built only for human visitors.
Where the index lands
The through-line of this edition is not that AI is failing. It’s that AI is being installed everywhere ahead of evidence that it earns its place — in search, in homes, in grocery supply chains, in cinema.
The report’s closing image is the strongest thing in it: like a modern Odysseus, you are cursed to roam AI-infested waters, dueling with machine gods and monsters, trying to find your way back to somewhere that feels like home.
Ten years after a machine beat the best Go player alive by searching smarter, the industry is still arguing about whether its replacement technology can search at all.
- TikTok Shop Launches Its Paid Membership Program in the U.S.
- Claroty Finds 65,000 Consumer Devices on Critical Infrastructure Networks
- SpaceX Share for Every Kid Needs a Charity and Treasury's Approval (daybreakwire.com)
- DRF Daily Playbook: Joseph R. Peluso Memorial Stakes Preview and Racing Analysis (archynewsy.com)