COLLISIONTechnologies

Intelligence needs experience.

LLMs trained mainly to imitate human text and judgements face a challenge on the path to superintelligence: their knowledge depends on what people have already produced.

In Welcome to the Era of Experience, David Silver and Richard Sutton argue that imitating human data alone is unlikely to deliver broadly superhuman capabilities. They propose agents that learn from their own actions and observed outcomes. [1]

World models offer a setting for that experience. Google DeepMind’s Genie 3 generates interactive worlds that respond to actions, giving agents environments to explore scenarios and consequences. [2]