The Stack Overflow Podcast “We’re not worried about compute anymore”: The future of AI models
Jun 13, 2025
Jamie DeGare, the Founding SVP of Product at Together AI, shares insights into the evolution of AI models and their future. He discusses the advantages of open-source AI and its role in enhancing research reproducibility. The conversation highlights innovations in AI infrastructure that offer cost-effective computing solutions. Jamie also addresses the critical aspects of data security and ethics when using open-source models, emphasizing the importance of safe data practices. Finally, he envisions a shift towards more comprehensive AI systems that integrate generative capabilities.
AI Snips
Chapters
Transcript
Episode notes
Future Is Many AI Models
- The future of AI models involves combining many specialized models rather than relying on a single "big model to rule them all."
- Using multiple models optimizes for cost, latency, and accuracy by assembling an agentic system.
Adopt Open Source for Control
- When moving beyond prototypes, organizations face challenges of cost, performance, and accuracy in using closed models.
- Use open source models and invest in expertise to gain better control through fine tuning and system performance improvements.
What Truly Makes A Model Open
- Open source models vary by openness: open weights, open training data recipes, and open training code.
- True reproducibility requires all three, but many users primarily need open weights to run models independently.