← Home
Practical AI #331 · September 1, 2025 · 48m
Episode 331: RAG and Reasoning — Whats Actually Working
Daniel and Chris discuss the current state of RAG (Retrieval-Augmented Generation) and reasoning models. Why many enterprise AI initiatives stumble and what architectures are actually working in production.
This summary was generated from show notes and public descriptions, not from a full transcript review. Details may contain inaccuracies.
Highlights
•
RAG is not a silver bullet — most RAG implementations fail because retrieval quality is poor, not because the generation model is wrong
Benson argues that enterprise RAG failures are usually retrieval failures: the system retrieves wrong or irrelevant documents, and the language model faithfully generates wrong answers from wrong context.•
The most underrated AI architecture for enterprise is fine-tuned small models, not prompted large models
Whitenack argues that for most enterprise use cases, a fine-tuned 7B parameter model outperforms a prompted GPT-4 at 1/100th the cost. The industry over-indexes on model size because it is easier to demo but harder to deploy.Was this useful?