Latest / Minds, Bodies, and Terawatts / Google's Chief Scientist Quit After 27 Years to Build a Company That Automates Scientists — and Google Is Funding It

Transcript

Google's Chief Scientist Quit After 27 Years to Build a Company That Automates Scientists — and Google Is Funding It · Minds, Bodies, and Terawatts

• 0:00
  [Host] This is Minds, Bodies, and Terawatts, where the last job anyone thought was safe just became a startup pitch. On Wednesday, August fifth, Google DeepMind lost its chief executive and Google lost its chief scientist, the same day.
• 0:14
  [Guest] Demis Hassabis left the DeepMind CEO seat for two new titles, Alphabet chief scientist and DeepMind chairman. Not a demotion. He's the name floated as Sundar Pichai's eventual successor.
• 0:25
  [Host] So who runs the lab tomorrow?
• 0:28
  [Guest] Koray Kavukcuoglu, the chief technology officer, now reports to Pichai and takes the Gemini models.
• 0:34
  [Host] That's the reshuffle. The exit is the story.
• 0:37
  [Guest] Jeff Dean left after twenty-seven years. He was Google's thirtieth employee, from the storefront days in Palo Alto.
• 0:44
  [Host] Why is this one different?
• 0:46
  [Guest] Because of what he left for. Dean and three longtime colleagues are founding Discovery Loop, a public benefit corporation. The stated mission, to automate machine learning, science and engineering.
• 0:57
  [Host] Automate research how?
• 0:58
  [Guest] Instead of a person running experiments one at a time, the system proposes an experiment, runs it, reads the result, designs the next, thousands at once.
• 1:08
  [Host] Pitch deck, or does it work?
• 1:10
  [Guest] DeepMind published the working version in May of twenty twenty-five, a paper called AlphaEvolve, a coding agent for scientific and algorithmic discovery. Its loop found a way to multiply two four-by-four complex matrices in forty-eight scalar multiplications, beating the forty-nine Strassen set in nineteen sixty-nine.
• 1:28
  [Host] Fifty-six years, closed by a search loop.
• 1:30
  [Guest] The same system rewrote a scheduling heuristic and handed Google back about zero point seven percent of its worldwide compute. Dean watched that run inside the building before leaving to sell it.
• 1:43
  [Host] So he quit to sell the automation of scientists.
• 1:46
  [Guest] He put it this way to the New York Times. We think there is opportunity for AI to more fully automate what has traditionally been a very human-intensive experimental loop.
• 1:58
  [Host] Where does Google land in that?
• 2:00
  [Guest] Founding investor, and the cloud supplier. Pichai's own statement called it a continuing partnership, not a break.
• 2:07
  [Host] Then this isn't a company losing its people.
• 2:10
  [Guest] Google moved its riskiest research off its own payroll, kept a stake, and bills the new company for the compute it runs on.
• 2:19
  [Host] The market read it as a loss anyway. CNBC reported Alphabet closed down about four percent.
• 2:25
  [Guest] No product failure, no bad earnings. Investors were pricing the people.
• 2:29
  [Host] Google has more compute, data and cash than almost anyone alive, and it couldn't hold them.
• 2:35
  [Guest] Once compute and capital go abundant, the bottleneck slides onto the few people who know what to build with them. Nobody can manufacture that input.
• 2:45
  [Host] And they walked out to build the thing that ends their own scarcity.
• 2:50
  [Guest] Their first market is machine learning research. Their lead investor said the founders are their own first customer.
• 2:57
  [Host] If you write research code for a living, the people who defined that job just told you where it goes.
• 3:05
  [Guest] The work that goes first is running experiments and tuning models, not deciding which question is worth asking.
• 3:12
  [Host] Somebody still has to read the output.
• 3:14
  [Guest] The AlphaEvolve authors named that limit themselves. It needs an automatic evaluator, so it only runs where the machine can score its own answer. Everywhere else, a person still reads.
• 3:26
  [Host] Patrick Deglon calls that the Substitution Threshold in Unscarcity, the point where the cheapest reliable provider of a job stops being human. Every earlier crossing was called by people whose own work was safe. This one came from the man who held the job.
• 3:43
  [Guest] For more on the talent bottleneck and what happens when the loop outruns its readers, head to unscarcity dot ai. The book is on Amazon, the audiobook on Spotify. This has been Minds, Bodies, and Terawatts.