Who is Demis Hassabis?

Demis Hassabis proposes the 'Einstein test' as a benchmark for AI novelty: train a system on knowledge up to 1901 and see if it independently derives Einstein's 1905 discoveries, including special relativity, to evaluate whether AI can generate genuinely new hypotheses rather than recombine known facts.

“Can you train a system with the knowledge of cutoff of 1901, and then will it come up with you know, what Einstein did in 1905, including special relativity, you know, his annus mirabilis.”

Demis Hassabis on the first visible sign of transformative AI: augmented humans at 1000x productivity, not autonomous agents.

“I don't think we'd see autonomous first. We would actually probably see people in this room operating at 1000X, and then That's what you should see first.”

AlphaFold: Demis Hassabis predicts it will be embedded in the discovery process of nearly every drug developed from this point forward.

“Almost every drug discovered from now on will have used AlphaFold at some point in its in the drug discovery process.”

Brian Greene is skeptical of 2029 AGI timelines put forward by Dario Amodei, Sam Altman, and Demis Hassabis, directly contradicting their forecasts.

“I'm skeptical that by 2029 this will really be the case.”

Dario Amodei on OpenAI's early motivation: fear that Demis Hassabis would seize control with AGI.

“The founders of OpenAI were talking back in like 2017 about how the reason why we made OpenAI was because we were worried that Demos at Google was going to become dictator with AGI.”

Demis Hassabis estimates AGI is 20 to 30 years away, a notably longer horizon than most public AI lab timelines.

“Mine's like 20 30 or something like this.”

Demis Hassabis estimates a full virtual cell simulation is approximately 10 years away.

“We're about 10 years away probably from something like a virtual cell, like a full virtual cell.”

Demis Hassabis puts the probability at roughly 50/50 that existing AI techniques can scale to AGI without any new fundamental ideas, with only one or two such ideas remaining if new ones are needed.

“I don't think it's more than one or two if there are out there. And I think, you know, my betting is about 50/50 if that's the case.”

Demis Hassabis argues that inference compute will NOT become essentially free, predicting chip production bottlenecks will force rationing for at least the next few decades.

“There'll be some bottleneck at least for the next few decades, I think. And so if that's the case, there'll still be rationing on the inference side.”

Demis Hassabis predicts AI will solve a Millennium Prize problem such as the Riemann hypothesis within approximately two years.

“Maybe we're a couple of years out from doing that.”