Mixture of Experts This week in AI models: Granite 4.0, Claude 4.5, Sora 2
19 snips
Oct 3, 2025 
Kaoutar El Maghraoui, a Principal Research Scientist, shares insights on model efficiency and why small models matter for enterprises. Kate Soule, Director of Technical Product Management for Granite, discusses the Granite 4.0 release and its roadmap, touching on governance certifications and future innovations. Kush Varshney, IBM Fellow in AI Governance, examines safety and transparency in AI models like Claude 4.5 and Sora 2, highlighting societal risks and the implications of AI-driven purchases. A lively discussion on the intersection of AI and cybersecurity wraps up the conversation.
AI Snips
Chapters
Transcript
Episode notes
Small Models, Big Impact
- Granite 4 emphasizes very small, efficient models that outperform prior larger versions while fitting on a single GPU.
- The release pairs efficiency gains with governance features like ISO 42001 to appeal to enterprises.
Transparency As A Differentiator
- IBM is pushing transparency in open-source model development with measures like cryptographic signing and engagement with the Stanford Transparency Index.
- These moves position Granite as a leader in governed open-source releases, not an outlier.
Efficiency Over Scale
- Granite reframes success from sheer capability to compute efficiency and practical enterprise needs.
- Efficiency reduces cost, regulatory exposure, and environmental impact while enabling consumer GPU deployment.