We're sunsetting PodQuest on 2025-07-28. Thank you for your support!
Export Podcast Subscriptions
cover of episode Google DeepMind CTO: Advancing AI Frontier, New Reasoning Methods, Video Generation’s Potential

Google DeepMind CTO: Advancing AI Frontier, New Reasoning Methods, Video Generation’s Potential

2025/5/20
logo of podcast Big Technology Podcast

Big Technology Podcast
AI Deep Dive AI Chapters Transcript
People
K
Koray Kavukcuoglu
Topics
Koray Kavukcuoglu: 我认为在AI模型的发展中，规模化和新技术是同等重要的。在生成式AI模型的研究中，架构和算法与规模同等重要。研究不同架构和算法在规模化下的效果至关重要。数据与算法、架构和计算能力同样关键。推理时间技术对于提升模型推理能力至关重要。我不认为扩大规模令人失望，因为我们一直在有效地推动模型的能力。重要的是进行广泛的研究，并从多个角度考虑扩展。各种模型都在显著改进，整个领域都在进步。Gemini模型在能力和质量上都取得了稳定进展。我们不断推进前沿技术，并在多个研究方向上看到回报，为实现AGI还有更多进展要做。

Deep Dive

Chapters
Koray Kavukcuoglu discusses the importance of scale versus novel techniques in advancing AI models. He highlights that scale is a significant factor, but other elements like data, algorithms, and inference-time techniques are equally crucial. The discussion touches upon the progress made in AI models despite diminishing returns from simply increasing scale.
  • Scale is important but not the only factor in advancing AI models.
  • Novel techniques, data, and inference-time improvements are equally crucial.
  • Diminishing returns from solely increasing scale are acknowledged, but progress continues through diverse research directions.

Shownotes Transcript

Koray Kavukcuoglu is the Chief Technology Officer of Google DeepMind. Kavukcuoglu joins Big Technology to discuss how his team is pushing the frontier of AI research inside Google as the company's Google IO developer event gets underway. Tune in to hear Kavukcuoglu break down the value of brute scale versus novel techniques and how the new inference-time “DeepThink” mode could supercharge reasoning. We also cover Veo 3’s sound-synced video generation, the open-source-versus-proprietary debate, and what a ten-percent jump in model quality might unlock for users everywhere.