Mixture of Experts LlamaCon, Qwen3, DeepSeek-R2 rumors and JP Morgan’s open letter on AI
May 2, 2025
Shobhit Varshney, Head of Data and AI for the Americas, Chris Hay, CTO of Customer Transformation, and Kush Varshney, an IBM fellow on AI governance, come together for a thrilling discussion. They dive into the buzz from LlamaCon and the implications of new AI models like Qwen3 and the rumored DeepSeek-R2. The trio also examines J.P. Morgan's urgent open letter on AI security, raising concerns about the balance between innovation and safety. Join them as they reflect on the past year’s AI journey and key moments from their debut conversation!
AI Snips
Chapters
Transcript
Episode notes
Meta's Lama API Platform
- Meta's Lama API centralizes development with fine-tuning and evaluation tools.
- It aims to combine open source flexibility with closed source stability for developers.
Multi-layered AI Safety Approach
- AI safety now involves multiple models across input, output, and firewall layers.
- Layered security reflects growing seriousness about AI protection and governance.
Hybrid AI Thinking Modes
- Hybrid "thinking" and "non-thinking" AI modes tailor response styles to tasks.
- Thinking modes simulate deliberation useful for reasoning, while non-thinking suits quick fact answers.