Big Technology Podcast Google DeepMind CTO: Advancing AI Frontier, New Reasoning Methods, Video Generation’s Potential
89 snips
May 20, 2025 
Koray Kavukcuoglu, the Chief Technology Officer of Google DeepMind, shares insights into cutting-edge AI research. He discusses the contrast between scaling AI efforts and embracing novel techniques, focusing on the new 'DeepThink' mode that enhances reasoning capabilities. Kavukcuoglu also highlights advancements in video generation, particularly with the VO3 model, which syncs sound and showcases complex object interactions. Furthermore, he addresses the ongoing debate between open-source and proprietary AI models, emphasizing the importance of responsible development for societal benefit.
AI Snips
Chapters
Transcript
Episode notes
Multi-Dimensional AI Progress
- Koray Kavukcuoglu emphasizes that progress in AI research equally requires scaling, architecture improvements, data quality, and inference-time techniques.
- He highlights that reasoning capabilities can be significantly enhanced by enabling repeated model use at inference time.
Optimism Beyond Scaling Limits
- Koray sees the diminishing returns of scale not as a disappointment but as a sign to pursue multiple research dimensions.
- He is optimistic about ongoing progress and believes there is still substantial room for improvements in AI capabilities.
AGI Needs Broad Innovations
- Koray stresses that reaching AGI requires critical innovations beyond just scaling large language models.
- He encourages maintaining a broad research agenda that considers multiple hypotheses to responsibly build AGI.