You will have solely just lately heard about Demis Hassabis. He’s been named one among Time journal’s “AI architects”, received a Nobel Prize for utilizing the know-how to foretell protein folding and runs Google’s AI efforts. When the search large acquired his firm DeepMind in 2014, he embraced his new employer’s huge sources to construct machines that surpassed human brainpower, so-called synthetic basic intelligence.
His greatest achievements since then have been to provide Google the glow of scientific status, with AI methods that beat the world’s high Go gamers and that Nobel Prize for chemistry. A product breakthrough has lengthy eluded Hassabis, however that would change in 2026 if his unconventional concepts make their manner into Google’s second try at sensible glasses, the sort of improvement that would mark a full turnaround for a corporation that was caught on the again foot three years in the past by ChatGPT.
Google plans to launch AI-powered sensible glasses subsequent yr to compete with related merchandise from Meta Platforms Inc. It’ll associate with Samsung to make the gadgets, one among which is able to include a small, in-lens show for issues reminiscent of navigation or translation. They may provide redemption to Hassabis’ firm in two methods. First, the unique Google Glasses appeared so absurd and labored so badly that they ruined the fame of sensible glasses for a number of years. Trendy, helpful AI specs would rehabilitate the know-how.
Second, it might show Hassabis’ long-standing perception that chatbots will not be the one path to highly effective, useful AI. OpenAI’s ChatGPT sparked the generative AI growth with massive language fashions educated on content material from the online and powered by lots of of billions of {dollars}’ price of computing energy. However Hassabis has held agency to the concept “world fashions” educated on simulations and the bodily world round us will result in the subsequent leap ahead.
Meta’s AI chief Yann LeCun thought the identical however couldn’t get Mark Zuckerberg to agree, and just lately left the corporate. Whereas Zuckerberg is chasing superintelligence by going all in on chatbots reminiscent of OpenAI, Google is hedging its bets by betting on scientists and potential paradigm-shifting applied sciences too.
For a begin, Sundar Pichai, the boss of Google guardian Alphabet, merged the corporate’s two AI models underneath the management of Hassabis in 2023, shocking many who thought the California-based engineering lead Jeff Dean would get the job. Hassabis refused to maneuver to Silicon Valley or undertake his employer’s go-to-market mindset, and he obsessed over scientific discovery. But from his base in London, he appeared to iron out the deep, trans-Atlantic rivalries between the 2 models that had stopped them progressing.
Then in August 2024, Google successfully spent $US2.7 billion to rehire Noam Shazeer, an eccentric scientific genius who co-invented the Transformer – a neural-network structure that varieties the spine of contemporary AI, and is successfully the T in ChatGPT. He’d stop Google in 2021 as a result of it refused to launch his chatbot program two years earlier than OpenAI shocked the world with the identical know-how. When Shazeer got here again final yr, he was made technical co-lead for the Gemini chatbot and reported to Hassabis, an association that on paper shouldn’t have labored out. Hassabis didn’t share Shazeer’s long-time obsession with LLMs as a path to AGI.
However Hassabis can also be a former chess prodigy who likes to assume 4 strikes forward, and he navigated his variations with Shazeer by using attraction and diplomacy. The stakes had grow to be greater. “He’s in a hyper-capitalist race that he didn’t begin, and now that he’s in it he’s decided to win,” says Sebastian Mallaby, writer of The Infinity Machine, a forthcoming e-book on Hassabis and DeepMind. “Though there’s a giant a part of his character that cares concerning the science there’s an equally vital half that cares about successful.”
He’s a good distance from victory. ChatGPT is effectively forward of Gemini on person numbers and has a two-year head begin on rolling out an enterprise enterprise, the all-important path to making a living from AI. (Gemini Enterprise solely launched in October.)
However Google’s newest AI mannequin was nonetheless a powerful success. Gemini 3 topped leaderboards on efficiency and prompted Sam Altman to name a code purple at OpenAI, citing “non permanent financial headwinds”. Gemini 3 now has greater than 650 million month-to-month app customers and an estimated 2 billion customers by way of AI Overviews on Google search. The corporate has all the time taken a gradual and cautious method to AI and now the tortoise is catching as much as the hare.
Shazeer deserves credit score for that comeback. Silicon Valley scuttlebutt has it that he managed to trace down a deep bug in Gemini, serving to to make it way more environment friendly to coach and to surpass ChatGPT on some benchmarks. Hassabis’ massive second continues to be to return. If he could make world fashions like his Challenge Astra work in apply, they might be the subsequent paradigm shift after chatbots: AI that understands how bodily objects transfer in house and relate to 1 one other.
Think about sporting glasses that keep in mind the place you left your keys, perceive what you’re in three dimensions and predict what may occur subsequent in your setting. That’d be a leap ahead from Meta’s Ray-Ban glasses, which may solely describe what they see by way of a digicam with out processing the physics of what’s round you. It might herald a step change in AI assistants.
The strain is mounting. Google should show it might monetise AI past adverts, and Hassabis wants one among his moonshots to lastly grow to be a viable enterprise. His observe file till now could be sobering. For all of the status of AlphaFold, the protein-structure predictor that’s accelerating the work of three million scientists, it has but to provide any FDA-approved medicine.
But when Google’s new glasses work and promote thanks partially to world fashions, that would put the corporate within the result in discover a killer app for AI. It can additionally decide whether or not Hassabis stays one among Google’s most embellished scientists, or turns into the architect of its subsequent period.
Bloomberg
Get information and evaluations on know-how, devices and gaming in our Know-how publication. Signal as much as obtain it each Friday.
Leave a Reply