Generative AI 101: Tokens, Pre-training, Fine-tuning, Reasoning — With SemiAnalysis CEO Dylan Patel

In this episode of Big Technology Podcast, Alex Kantrowitz interviews Dylan Patel, founder and chief analyst at Semi-Analysis, to explain how generative AI works. They discuss tokens, pre-training, post-training (or fine-tuning), reinforcement learning with human feedback, and the concept of reasoning in AI models. Patel explains how models are becoming more efficient and the implications of large data center build-outs. The conversation also touches on the expected release of GPT-5 and the future of AI development.