Korva is a scroll of the internet’s best ideas.
Discovery Loop launches
Discovery Loop will focus on deep scientific breakthroughs in AI. Dean joined Google in 1999 as employee number 30 and has worked there continuously for 27 years. The 4% drop in Google's shares equals about $200 billion in lost market value.
Hassabis moves up
The departure came on the same day as a second shakeup. On Wednesday, Demis Hassabis moved to chair of DeepMind and of Google. Reports describe the move as stepping down or being kicked upstairs, while Google framed it as a promotion.
Gemini delays and spending
Axios reported that Gemini 3.5 Pro is months behind, with some company sources attributing the delay in part to low morale. Several top researchers, including Gemini's co-lead, have left for competing AI labs. Google has committed to $200 billion in capital spending this year on AI infrastructure and data centers.
Why can one departure move a stock?
A share price reflects what investors expect a company to earn in the future, and a company's best researchers are part of that expectation. Market value is the share price times the number of shares, so a 4% move in a company worth trillions is hundreds of billions of dollars. Jeff Dean co-created MapReduce and Bigtable, two systems Google described in research papers in the mid-2000s.
How did Jeff Dean shape Google's infrastructure?
Dean worked with Sanjay Ghemawat on systems that let Google run searches and processing jobs across thousands of ordinary machines. Their 2004 paper described MapReduce, a programming model for splitting a huge job across many computers, and it inspired the open-source Hadoop project. They also built Bigtable, a storage system described in a 2006 paper, and Spanner, a globally distributed database described in a 2012 paper. Dean later helped start Google Brain, the research team formed around 2011, and co-created the TensorFlow software library.
Why do researchers leave big labs to found startups?
Large companies decide what to build based on products and revenue, while a startup lets researchers pick one problem and own the result. The best-known example is the 2017 Google paper Attention Is All You Need, written by eight researchers, which introduced the transformer architecture behind today's language models. Several of its authors later founded or joined startups, including Cohere and Character.AI.
More from All-In with Chamath, Jason, Sacks & Friedberg
