The Stack Overflow Podcast “The future is agents”: Building a platform for RAG agents
60 snips
May 27, 2025 
Douwe Kiela, CEO of Contextual AI and a pioneer in retrieval augmented generation, shares insights on building platforms for RAG agents. He delves into the evolution of AI and its impact on natural language processing. A key topic is addressing the hallucination issue in generative models, showcasing how RAG enhances accuracy. Douwe also discusses the role of synthetic data in model training and the future possibilities of intelligent agents, emphasizing the transformative potential of these technologies.
AI Snips
Chapters
Transcript
Episode notes
Douwe Kiela's Academic Journey
- Douwe Kiela started interested in technology young but studied philosophy before computer science.
- He switched to computer science in grad school and did a PhD at Cambridge, focusing on NLP and AI's early days.
Origins of Retrieval-Augmented Generation
- RAG emerged from combining existing components like vector databases and language models around 2020-2021.
- It enables grounding language models in flexible external knowledge to reduce hallucinations.
Hallucinations Versus Mistakes
- Hallucination means making up knowledge not in the ground truth, distinct from regular errors.
- RAG decouples knowledge retrieval from generation to control and reduce hallucinations.