To Survive the AI Era, You Need a New Metric: Founder-Testing Velocity

Demis Hassabis, the head of Google DeepMind, stood in the shadow of the Acropolis and gave a lecture on the future. He said the key to survival in the AI era is “meta-learning”; learning how to learn.

It’s a smart take from a brilliant guy. The new documentary about his work, The Thinking Game, tracks DeepMind’s incredible journey from playing board games to winning a Nobel Prize for solving protein folding.

It portrays a team relentlessly pursuing scientific perfection.

But there is a massive irony here.

Despite DeepMind’s emphasis on “meta-learning” and their unparalleled research, they weren’t the ones who triggered the global “AI breakthrough” that changed our daily lives.

OpenAI earned that title.

And they didn’t do it by staying in the lab until they had AGI. They did it by doing the exact opposite: gradually bringing a “less-deficient” product to market.

OpenAI released GPT-1, then 2, then 3. They were flawed. They hallucinated. They were imperfect. But by shipping these “deficient” products, they gathered the data and user feedback that allowed them to build ChatGPT.

While DeepMind was playing The Thinking Game, OpenAI was playing The Shipping Game.