AlphaGo builder's new London lab bets LLMs can't be trusted to reason
Thore Graepel, who helped build DeepMind's AlphaGo, is applying game-winning search techniques to general reasoning. His warning: LLMs can't be trusted to reason.

Loading the Newsroom
Curating from trusted global sources…
13 briefings · “wan”
Thore Graepel, who helped build DeepMind's AlphaGo, is applying game-winning search techniques to general reasoning. His warning: LLMs can't be trusted to reason.
A walkout at one of the world's key memory chip hubs threatens supply just as AI demand peaks, and puts Taiwan's profit-sharing norms to the test.
The journalist who exposed Theranos says the Oscar-buzzed film is art, not evidence, and the distinction matters for anyone weighing a founder's story.
Tech leaders want the shares given directly to Americans, not parked in a foundation.
Alan Milburn's call for welfare cuts to fund employment support opens a Labour fight over benefits and the Neet crisis.
Woodside's Scarborough gas project is frozen mid-start-up after activists boarded the platform, a disruption with cost and reputational ripple effects for LNG operators.
The researcher's new book argues standardized AI tests are fundamentally broken, but he says they can be rebuilt into a real science.
The biggest US bank just made agentic AI's consumer moment a security question, and Meta's finance push is caught in the crossfire.
Bhusri is shifting Workday to usage-based AI pricing, targeting the tens of millions customers spend on outsourcing.
David Ellison says the Paramount-WBD merger has full clearance after settling with state AGs, clearing the path for an October 1 close.
A Nasdaq opening pop turns SpaceX into a public company overnight, forcing new money questions fast.
The ChatGPT maker’s filing joins Anthropic’s AI listing wave and turns “someday” into a real timing option.