Lightcone Podcast How To Build Generative AI Models Like OpenAI's Sora
Apr 30, 2024
Discover how foundational AI models can be built in under three months, challenging the notion that only big companies can succeed. Learn about innovative advancements in generative AI and impressively realistic video modeling, including OpenAI's Sora. Exciting developments in lip-sync technology and synthetic data show how startups are revolutionizing the field. Plus, hear about unique projects from Y Combinator's Winter 24 batch, emphasizing creativity and accessibility for aspiring AI founders.
AI Snips
Chapters
Transcript
Episode notes
Infinity AI's Deepfake Podcast
- Infinity AI created deepfake podcast videos using just YouTube episodes.
- They adapted a foundation model, needing only an hour of footage.
SyncLab's Lip-Syncing
- SyncLab created real-time lip-syncing on a single A100, using low-res video.
- YC's Azure GPU cluster enabled 100x faster iteration.
Sonato's Text-to-Music
- Sonato, built by 21-year-old college grads, generates custom songs from lyrics.
- It's one of the few, and arguably best, text-to-music models.