Demis Hassabis - Scaling, Superhuman AIs, AlphaZero atop LLMs, Rogue Nations Threat
Dwarkesh Podcast
Embracing Multi-Modality in Robotics and Transfer Learning
The advancements in robotics like Gatto and RT2 are driven by the data pool regime enabling research in various directions such as sampling efficiency, data efficiency, transfer learning, and learning from simulation. The transfer of large models to robotics is becoming feasible, treating different tokens like actions, words, or pixels as part of a true multi-modality approach. Training such multi-modal systems is initially challenging but beneficial as understanding different modalities enhances performance in each domain.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
