Hello Betamax, Plenty of people learn to cook from YouTube. Watch enough videos of someone making xiaolongbao (soup dumplings) and eventually, something clicks. In that context, the cooking technique is learned through sheer repetition. Whether that counts as understanding or just good mimicry is a question most home cooks wouldn't bother asking. Can the same be said about AI? Our top story today looks at Video Rebirth, a Singapore-based startup building world models - AI models that understand the dynamics of the real world - by training them on videos. Co-founder and CEO Wei Liu believes physical rules can be implicitly learned rather than pre-programmed. While the goal is to have world models that can simulate and predict how physical reality behaves, Video Rebirth's current business focuses on video generation for enterprises, either through API access or model customization for companies that want to differentiate using their own intellectual property. Video Rebirth is far from the only firm pursuing world models. Runway, World Labs, and AMI Labs all have some version of a world model ambition and have secured billions of dollars of funding in the process. The space has the feel of the early large language model days: real potential, real money, and expectations that often run well ahead of what the technology can do. For now. Glenn Kaonang, journalist |
