i
News
News · 2026-10-21

Demis Hassabis puts AGI two to three years away

@neuronium_ai @neuronium_ai

Demis Hassabis now puts the arrival of artificial general intelligence two to three years away, closer than his earlier estimate this year of three to five. In an interview with The Wall Street Journal, the Google DeepMind co-founder said his forecast keeps moving toward the present. His proposed test is not whether a model can reproduce existing knowledge, but whether it can make a scientific leap: could a system trained only on knowledge available before 1901 independently recreate Einstein’s 1905 papers?

Cover: Demis Hassabis puts AGI two to three years away

A forecast that keeps moving

Hassabis has compared AGI’s impact with the Industrial Revolution, but ten times larger and unfolding ten times faster. The scale of that claim makes the shrinking timeline consequential; the interview offers a test for capability, but no evidence that a system is close to passing it.

After leaving his post at Google DeepMind, Hassabis has focused on Isomorphic Labs, a company spun out of Google that applies AI to drug discovery.

The gap between prediction and preparation

On safety, Hassabis has recently proposed an industry-funded oversight council under government control. It would assess models before release. He has warned that bad regulation is worse than no regulation.

I think the most revealing detail is what remains unmeasured: the Einstein test is a vivid benchmark, but the interview gives no result against it. A shorter forecast is not the same as demonstrated progress, and the proposed oversight council is still a proposal. Hassabis is asking the public to take both the pace of progress and the need for oversight seriously before the central capability has been shown.

Daily AI news

Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.

Only what matters — every day

Follow on X