The best machine learning podcasts are Built This Week, The TWIML AI Podcast, Gradient Dissent, Practical AI, Latent Space, Dwarkesh Podcast, The Cognitive Revolution and Lex Fridman Podcast. TWIML and Gradient Dissent are the most ML-specific; Latent Space covers AI engineering; Dwarkesh and Lex Fridman go long with the researchers themselves.
Built This Week is our show, so we put it first; here's how we picked the rest.
Built This Week is hosted by Jordan Metzner and Sam Nadler, co-founders of Ryz Labs, weekly since June 2025, with 53 episodes so far. To be upfront, it is not an ML research show: each episode we build something with AI on air, test new AI tools, interview a founder or operator, and cover the AI news, in about 21-28 minutes. It belongs on this list as the applied end, for ML people who want to see how models get used in products and what the hardware side looks like.
For example, with Positron CEO Mitesh Agrawal Sam built an AI inference cost simulator comparing GPUs and Positron's chips, and Episode 50 covered an AI physics model for chip design. Both feed our AI in semiconductors page.
Start with: Episode 51: It Ships While You Sleep (September 2026), with David Petrou, who spent almost 18 years at Google, on splitting work across coding agents. See all episodes.
Sam Charrington has hosted TWIML, formerly This Week in Machine Learning & AI, since 2016. It publishes roughly every one to two weeks, with interviews of ML and AI researchers and practitioners on machine learning, deep learning, NLP and applied AI, typically in 55-66 minutes. It is aimed squarely at ML researchers, data scientists and engineers, and it is rated 4.7 out of 5 from 426 ratings on Apple Podcasts, as of October 2026.
Start with: Inside Nano Banana 🍌 and the Future of Vision-Language Models with Oliver Wang (September 2025). Read our full profile of The TWIML AI Podcast.
Gradient Dissent comes from Weights & Biases, whose founder Lukas Biewald interviews AI leaders and practitioners about the newest AI work and about bringing models into production. New episodes arrive roughly every two to four weeks and typically run 50-68 minutes. It is a natural fit for ML engineers and AI leaders who care about production ML and infrastructure; it is rated 4.8 out of 5 from 70 ratings on Apple Podcasts, as of October 2026.
Start with: Exploring PyTorch and Open-Source Communities with Soumith Chintala, VP/Fellow of Meta, Co-Creator of PyTorch (July 2023). Read our full profile of Gradient Dissent.
Daniel Whitenack and Chris Benson have hosted Practical AI weekly since 2018, mixing discussion and guest interviews about making AI practical, productive and accessible. It emphasizes real-world implementations across ML, MLOps and LLMs, in episodes of about 45-50 minutes. It suits practitioners who want grounded, implementation-focused discussion.
Start with: Open models and the future of Physical AI with NVIDIA (October 2026). Read our full profile of Practical AI.
Latent Space is a podcast "by and for AI engineers," led by swyx with co-hosts Alessio and Vibhu. One to two episodes a week cover foundation models, code generation, agents and GPU infrastructure through builder interviews plus news and paper discussion. Episodes typically run 69-101 minutes, and it suits engineers who build on top of models every day.
Start with: ⚡️Satya Nadella: No Priors x Latent Space Crossover Special at Microsoft Build (June 2026). Read our full profile of Latent Space: The AI Engineer Podcast.
Dwarkesh Patel's deeply researched long-form interviews focus on AI, science and history, including many frontier-lab leaders and AI researchers. It publishes roughly weekly to every few weeks, and episodes typically run 76-133 minutes. It is for listeners who want detailed conversations about AI progress, scaling and research directions; it is rated 4.5 out of 5 from 615 ratings on Apple Podcasts, as of October 2026.
Start with: Ilya Sutskever — We're moving from the age of scaling to the age of research (November 2025). Read our full profile of Dwarkesh Podcast.
Nathan Labenz interviews AI builders and researchers about twice a week and publishes "AI Scouting Reports" and news segments on AI's near-term impact. Episodes typically run 102-141 minutes, with room for detail on capabilities, safety and policy. It suits researchers and AI-curious professionals who like long, careful conversations.
Start with: Dean Ball, on Joining OpenAI: New Power Centers, Frontier AI Policy, & Main Character Energy (June 2026). Read our full profile of The Cognitive Revolution.
Lex Fridman's independent show runs very long conversations, typically 174-280 minutes, released about every two to three weeks recently. Topics span AI, programming, science, history and politics, and the guests include prominent AI researchers and founders. It is for listeners who want multi-hour conversations with the people building the field; it is rated 4.7 out of 5 from 12.2K ratings on Apple Podcasts, as of October 2026.
Start with: #475 – Demis Hassabis: Future of AI, Simulating Reality, Physics and Video Games (July 2025). Read our full profile of Lex Fridman Podcast.
More focused on shipping software than on ML? See the best AI podcasts for developers, or the full list of the best AI podcasts.
The TWIML AI Podcast (formerly This Week in Machine Learning & AI) has interviewed ML researchers and practitioners since 2016. Gradient Dissent from Weights & Biases focuses on bringing models into production, and Practical AI on real-world implementations.
Dwarkesh Podcast and Lex Fridman Podcast host long conversations with frontier-lab leaders and researchers, including Ilya Sutskever, Andrej Karpathy, Dario Amodei and Demis Hassabis.
Gradient Dissent, hosted by Weights & Biases founder Lukas Biewald, covers bringing models into production. Practical AI also covers MLOps alongside ML and LLMs.
