1 hour ago
5
ARTICLE AD BOX
Demis Hassabis secretly attempted to launch a high-frequency trading hedge fund within DeepMind, aiming to rival Jim Simons' Renaissance Technologies. Google, however, shut down the project. This endeavor was part of DeepMind's broader, ultimately unsuccessful, efforts to gain independence from Google's control, leading Hassabis to believe in influencing decisions from within.
The team of ~20 researchers was training high-frequency trading algorithms—and Hassabis reportedly wanted to take on Jim Simons.Demis Hassabis once tried to build a hedge fund inside DeepMind.
Google shut it down. According to The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence by Sebastian Mallaby—excerpted this week in Colossus magazine—Hassabis quietly assembled a team of around 20 researchers to develop high-frequency trading algorithms, without Google's approval. When the parent company found out, the project was disbanded.The target Hassabis had in mind was Jim Simons. Specifically, he wanted to beat Renaissance Technologies, the notoriously secretive quant fund that Simons founded on Long Island.
The appeal was obvious: Renaissance ran like a classified operation, staffed by scientists, answering to no one. Hassabis—a five-time World Games Champion at the Mind Sports Olympiad—saw it as a game worth playing.
The man he wanted to beat slept next to his desk
Renaissance's longtime leader Peter Brown was a deep learning pioneer who had studied under Geoffrey Hinton. He kept a fold-down bed propped against his office wall and, by most accounts, lived at the office.
Hassabis recognized a kindred obsessive—a scientist who had turned a cloistered, secretive operation into the most successful hedge fund in history. That was the benchmark. Google's wariness killed the project before it got anywhere close.
Demis Hassabis' secret hedge fund was one of many times Google pulled DeepMind back
The trading team sits inside a much larger story that Mallaby's book excavates in detail: DeepMind's years-long, repeatedly failed attempt to break free from Google's grip.Hassabis and co-founder Mustafa Suleyman spent the better part of three years pushing for a governance structure they called Project Mario—a plan to spin DeepMind out as an independent "global interest company" with a 3-3-3 oversight board. Reid Hoffman committed $1 billion to back a potential walkaway. It never came together. Google CEO Sundar Pichai kept the door open long enough that the talks dragged on, then closed it.By the time it was over, Hassabis had drawn a different conclusion about AI safety. Negotiating trustless governance structures in advance, he told Mallaby, was a dead end. The lines always get drawn in the wrong places. What actually matters, he now believes, is staying inside the room where decisions get made—and earning trust over time.Whether that's wisdom or rationalisation is the question Mallaby's book leaves hanging.
English (US) ·