The bitter lesson teaches us that simple techniques at scale are the most effective. This has been proven with LLMs and has similarly shown true in video models, where scale has unlocked truly emergent behaviors in 3D and is showing early glimpses of the future of multimodal intelligence.
We’ll discuss the approach to, and findings of training Dream Machine, Luma’s SOTA video generation model.
Amit Jain is the CEO and Co-Founder of Luma AI, a startup pursuing visual intelligence through multi-modal foundation models.
Luma is working on natively combining the next generation of AI with delightful products, and most recently released Dream Machine, a market-leading Video Generation model that garnered 1MM users in 4 days.
Amit founded Luma from Apple, where he worked as a Systems and Machine Learning Engineer, leading work on the Passthrough feature for the Apple Vision Pro and before that on the integration of the first LIDAR sensors for the iPhone.
Amit also has experience in multiple startups, and studied Mathematics and Physics at university.