Overview
- Speaking at Axios' AI+ Summit in San Francisco, Demis Hassabis said current systems should be scaled "to the maximum" and suggested scaling could constitute the entirety of a future AGI.
- Hassabis added that achieving AGI will likely also require one or two additional breakthroughs beyond raw scale.
- The stance follows DeepMind’s widely praised release of Gemini 3, which the company presents as evidence of progress from scaling.
- Coverage highlights practical constraints on scaling, including finite public training data, high costs, and environmental impacts from building additional data centers.
- Yann LeCun argues many important problems scale poorly and is leaving Meta to launch a startup focused on "world models" that emphasize physical understanding, memory, reasoning, and planning.