In 2016, a Bloomberg reporter noticed that nearly every AI paper seemed to use Nvidia GPUs. He thought there was a major story there and tried to write it. The story never ran.
Ten years later, that reporter was a co-founder of Anthropic.
His name is Jack Clark. I recently came across an interview with him, and the more I read, the more I felt this was a story every technology journalist should know.
At Bloomberg, Jack covered Google and AI. For two years, he printed large numbers of research papers from arXiv and charted advances in computer vision, language, video and Atari games.
He noticed something striking: capability curves across these fields appeared to be accelerating at the same time. Nvidia GPUs, meanwhile, seemed to be everywhere.
On August 2, 2016, he messaged his Bloomberg colleague Joe Weisenthal. He said he planned to leave journalism, spend a few months properly studying AI, and then find work connected to it.
His reason came down to one thought:
“AI is more important than anything else, so I felt I should organize my life around it.”
When his editor, Alistair Barr, heard that Jack wanted to join a nonprofit research organization called OpenAI, he thought it was a terrible decision. OpenAI was less than a year old. Even its long-term funding was uncertain.
Jack left anyway. He later became OpenAI’s policy lead and, in 2021, co-founded Anthropic with Dario Amodei and others.
Jack had no secret inside information. The papers were public. The technical results were public. The use of Nvidia GPUs in those studies was hardly a secret.
What he did differently was put scattered information into a timeline. He saw signs of several fields advancing together, then trusted his own research enough to act on it.
That story makes me think about AI coverage today.
Every day, we follow new models, funding rounds and products. We rush to decide which company is winning and which product will disrupt another. The news moves so quickly that it is easy to mistake keeping up with it for being ahead of it.
But take away the daily headlines. How much independent judgment do we really have about where AI will be in five years? And how much of that judgment comes from research we have done ourselves?
When information becomes cheap, the scarce skill may be recognizing the importance of something everyone can already see.
Jack Clark did that ten years ago. His story offers technology journalists a lesson: sustained research can reveal something more valuable than the next headline.
He could not persuade Bloomberg to publish his Nvidia story. In the end, he used the same research to persuade himself to leave.
Looking back, perhaps the most consequential story he chose was the direction of his own life.
Comments