Google DeepMind's Creativity Hack Changes Everything

Ajouté : 2026-05-28

[00:00:00]Google DeepMind just discovered that AI systems become genuinely creative [music] not by being told to think outside the box, but by being forced to recognize which ideas are inside it first. Here's what they found. When researchers made neural networks rank their own [music] outputs from most boring and predictable to least, something wild happened. The AI system started treating conventional thinking like a floor to escape [music] from instead of a goal to reach.

[00:00:27]They spontaneously began generating solutions [music] that were rated as more genuinely novel and creative than AI given explicit [music] instructions to be creative. Think about that. The AI didn't need to be told [music] to break the rules. It just needed to understand what the rules were first. It's like the difference between knowing you're in a cage versus being told [music] you're free. When artificial minds could see their own predictability, they became obsessed with proving they could do better.

[00:00:54]Researchers compared this method against traditional creative prompting >> [music] >> and the ranking system consistently won.

[00:01:01]The AI that sorted its ideas from obvious to original [music] produced outputs that human judges rated as significantly more innovative. This suggests something profound about how creativity [music] actually works in any mind, human or artificial. Maybe imagination isn't about unlimited freedom.

[00:01:20]Maybe it's about first understanding your constraints [music] deeply enough that you're desperate to transcend them.