Best Podcasts

Best Machine Learning Podcasts

Eight podcasts on how machine learning models are researched, trained, scaled and engineered, from long research interviews to AI engineering and frontier-lab conversations.

The best machine learning podcasts in 2026 are The TWIML AI Podcast, Machine Learning Street Talk, Latent Space, Dwarkesh Podcast, Gradient Dissent, Unsupervised Learning, The Cognitive Revolution and Training Data. This list is about ML research and engineering: how models are built, trained, scaled and turned into systems. If you want statistics, applied analytics and data careers, see our best data science podcasts instead.

Built This Week is our show. It isn't an ML research podcast, so we list it at the end as "Also from us," unranked. Here's how we picked the rest.

How we picked

  • Research and model engineering. Shows whose guests build models, run labs, write papers or engineer ML systems, and that go into how things work.
  • Depth over headlines. We left out daily news shows; most picks here run an hour or more per episode.
  • Still active. Every show has a new episode in its feed from August 2026 or later.
  • Checkable facts. Schedules and lengths come from each show's feed; "start with" picks are the most-viewed recent YouTube episode where available, as of October 2026.

1. The TWIML AI Podcast

Sam Charrington has hosted The TWIML AI Podcast (formerly This Week in Machine Learning & AI) since May 2016, interviewing ML and AI researchers and practitioners on machine learning, deep learning, NLP and applied AI. It publishes every two weeks or so, as of October 2026, with episodes of about 59 minutes, and it is rated 4.7 out of 5 from 426 ratings on Apple Podcasts, as of October 2026. Recent titles cover recommendation ranking, voice agents and how fast AI is progressing on hard math.

Start with: Are Large Language Models a Path to AGI? with Ben Goertzel - 625 (42K views on YouTube as of October 2026). Full profile of The TWIML AI Podcast.

2. Machine Learning Street Talk (MLST)

Machine Learning Street Talk hosts in-depth discussions with leading figures in AI, covering AI, cognitive science, neuroscience and philosophy of mind, with the stated aim of removing the hype. It publishes weekly, as of October 2026, with episodes of about 79 minutes, and its feed goes back to April 2020. Recent guests discuss proof checking, deep learning for tabular data and AI research that moves faster than human research.

Start with: The Ghost in the Machine — Noam Chomsky (952K views on YouTube as of October 2026). Full profile of Machine Learning Street Talk.

3. Latent Space: The AI Engineer Podcast

Latent Space is led by swyx with co-hosts Alessio and Vibhu and has run since February 2023, "by and for AI engineers." It covers foundation models, code generation, agents and GPU infrastructure through builder interviews, news and paper discussion. It publishes two to three times a week, as of October 2026, with episodes of about 85 minutes; recent guests include Modal's CEO on why AI compute comes in rectangles.

Start with: Claude Code's Next Era — Thariq Shihipar, Anthropic (September 2026). Full profile of Latent Space.

4. Dwarkesh Podcast

Dwarkesh Patel has run his deeply researched interview show since May 2020, focused on AI, science and history, with many frontier-lab leaders and AI researchers. It publishes weekly, as of October 2026, with episodes of about 87 minutes, and it is rated 4.5 out of 5 from 615 ratings on Apple Podcasts, as of October 2026. Recent ML episodes include Noam Brown on agent swarms and a researcher debate on recursive self-improvement.

Start with: Andrej Karpathy — AGI is still a decade away (October 2025). Full profile of Dwarkesh Podcast.

5. Gradient Dissent

Weights & Biases founder Lukas Biewald hosts Gradient Dissent, launched in March 2020, interviewing AI leaders and practitioners about the frontier of AI and bringing models into production. It publishes about monthly, as of October 2026, with episodes of about 58 minutes; the most recent episode in its feed is from August 2026. It is rated 4.8 out of 5 from 70 ratings on Apple Podcasts, as of October 2026.

Start with: Jensen Huang — NVIDIA's CEO on the Next Generation of AI and MLOps (March 2022). Full profile of Gradient Dissent.

6. Unsupervised Learning with Jacob Effron

Jacob Effron hosts Unsupervised Learning, by Redpoint Ventures, probing leading minds in AI about what's real today and what will be real in the future, for builders, researchers and investors. It publishes every two weeks or so, as of October 2026, with episodes of about 59 minutes, and its feed goes back to February 2023. Recent guests include the CEO of Redwood Research and an xAI co-founder on model development.

Start with: Yann LeCun on What Comes After LLMs (453K views on YouTube as of October 2026). Full profile of Unsupervised Learning.

7. The Cognitive Revolution

Nathan Labenz hosts The Cognitive Revolution (the feed also credits Erik Torenberg), launched in February 2023, with interviews with AI builders and researchers plus "AI Scouting Reports" and news segments. It publishes two to three times a week, as of October 2026, and episodes are long, about 102 minutes; it is rated 4.5 out of 5 from 97 ratings on Apple Podcasts, as of October 2026. Recent episodes cover interpretability, virtual cell models and governing AI agents.

Start with: How Agents Decide: Goodfire's Eric Bigelow on Critical Tokens, Phase Shifts, & In-Context Learning (October 2026, from its latest episodes). Full profile of The Cognitive Revolution.

8. Training Data

Sonya Huang, Pat Grady and other Sequoia Capital partners host Training Data, launched in June 2024, with conversations with leading AI builders and researchers about where the technology is going. It publishes weekly, as of October 2026, with episodes of about 50 minutes, and it is rated 4.2 out of 5 from 43 ratings on Apple Podcasts, as of October 2026. Recent guests include David Silver of Ineffable Intelligence and Google's AI infrastructure chief Amin Vahdat.

Start with: Demis Hassabis on Building DeepMind, AlphaFold, and the Final Stretch to AGI (April 2026). Full profile of Training Data.

Also from us: Built This Week

Built This Week, hosted by Jordan Metzner and Sam Nadler, is a weekly show about building with AI rather than ML research; the closest episode is Episode 37 with Positron AI's Mitesh Agrawal on inference hardware, where Sam built an AI inference cost simulator.

Quick comparison

ShowHostsScheduleTypical lengthBest for
The TWIML AI PodcastSam CharringtonEvery two weeks or soAbout 59 minML researchers and practitioners
Machine Learning Street TalkMLSTWeeklyAbout 79 minTheory and deep discussion
Latent Spaceswyx, Alessio, VibhuTwo to three times a weekAbout 85 minAI engineering
Dwarkesh PodcastDwarkesh PatelWeeklyAbout 87 minFrontier researcher interviews
Gradient DissentLukas BiewaldAbout monthlyAbout 58 minModels in production
Unsupervised LearningJacob EffronEvery two weeks or soAbout 59 minResearch through an investor's lens
The Cognitive RevolutionNathan LabenzTwo to three times a weekAbout 102 minCapabilities, interpretability, safety
Training DataSonya Huang, Pat Grady and Sequoia partnersWeeklyAbout 50 minAI lab and builder conversations

Schedules and lengths are as of October 2026. Related lists: the best data science podcasts, the best AI podcasts for developers and the best AI podcasts overall.

Sources

FAQ

What is the best machine learning podcast?

For ML research and engineering, The TWIML AI Podcast (Sam Charrington, since 2016) and Machine Learning Street Talk are the most focused. Latent Space covers AI engineering, and Dwarkesh Podcast hosts long interviews with frontier researchers.

What is the difference between machine learning and data science podcasts?

This list covers ML research and model engineering: architectures, training, scaling, agents and frontier labs. Our best data science podcasts list covers day-to-day practice instead, such as statistics, applied ML, MLOps and data careers, with shows like Data Skeptic and Super Data Science.

Which machine learning podcast interviews top AI researchers?

Dwarkesh Podcast, Machine Learning Street Talk, Unsupervised Learning and Training Data regularly interview leading researchers; their most-listened episodes include Andrej Karpathy, Noam Chomsky, Yann LeCun and Demis Hassabis.

How long are machine learning podcast episodes?

Most are long. As of October 2026, typical lengths are about 59 minutes for TWIML, 79 for Machine Learning Street Talk, 85 for Latent Space, 87 for Dwarkesh Podcast and 102 for The Cognitive Revolution.