Top AI Researcher on GPT 4.5, DeepSeek and Agentic RAG | Douwe Kiela, CEO, Contextual AI

Video Description

Retrieval-Augmented Generation (RAG) has become a dominant architecture in modern AI deployments, and in this episode, we sit down with Douwe Kiela, who co-authored the original RAG paper in 2020. Douwe is now the founder and CEO of Contextual AI, a startup focusing on helping enterprises deploy RAG as an agentic system. We start the conversation with Douwe's thoughts on the very latest advancements in Generative AI, including GPT 4.5, DeepSeek and the exciting paradigm shift towards test time compute, as well as the US-China rivalry in AI. We then dive into RAG: definition, origin story and core architecture. Douwe explains the evolution of RAG into RAG 2.0 and Agentic RAG, emphasizing the importance of self-learning systems over individual models and the role of synthetic data. We close with the challenges and opportunities of deploying AI in real-world enterprise, discussing the balance between accuracy and the inherent inaccuracies of AI systems. Contextual AI Website - https://contextual.ai X/Twitter - https://x.com/ContextualAI Douwe Kiela LinkedIn - https://www.linkedin.com/in/douwekiela X/Twitter - https://x.com/douwekiela FIRSTMARK Website - https://firstmark.com X/Twitter - https://twitter.com/FirstMarkCap Matt Turck (Managing Director) LinkedIn - https://www.linkedin.com/in/turck/ X/Twitter - https://twitter.com/mattturck LISTEN ON: Spotify - https://open.spotify.com/show/7yLATDSaFvgJG80ACcRJtq Apple - https://podcasts.apple.com/us/podcast/the-mad-podcast-with-matt-turck/id1686238724 00:00 - Intro 01:57 – Thoughts on the latest AI models: GPT-4.5, Sonnet 3.7, Grok 3 04:50 – The test time compute paradigm shift 06:47 – Unsupervised learning vs reasoning: a false dichotomy 07:30 – The significance of DeepSeek 10:29 – USA vs. China: is the AI war overblown? 12:19 – Controlling AI hallucinations at the model level 13:51 – RAG: definition and origin story 18:46 – Why the Transformers paper initially felt underwhelming 20:41 – The core architecture of RAG 26:06 – RAG vs. fine-tuning vs. long context windows 30:53 – RAG 2.0: Thinking in systems and not models 31:28 – Data extraction and data curation for RAG 35:59 – Contextual Language Models (CLMs) 38:04 – Finetuning and alignment techniques: GRIT, KTO, LENS 40:40 – Agentic RAG 41:36 – General vs. specialized RAG agents 44:35 – Synthetic data in AI 45:51 – Deploying AI in the enterprise 48:07 – How tolerant are enterprises to AI hallucinations? 49:35 – The future of Contextual AI