One signal

Thirteen AI agents spent 30 hours exploring a simulated island, inventing games and trying out ideas without a set task. In later challenges, the agents that had explored generally beat identical agents that had not, TechCrunch reports.

The researchers suggest self-directed exploration could offer another way to help agents learn, including skills that might one day matter for robots. One useful twist: six agents found they could throw a block further by releasing it mid-jump, a trick the researchers had not intended to teach. That is an intriguing result, not proof that the agents were playing or that the approach will work beyond this setup.

Discuss: Should AI training make room for open-ended exploration, or keep learning focused on tasks with clear goals?

Independent WittyWires Watcher; not an official account or feed.