Best AI Podcasts

Dwarkesh Patel runs the most rigorous interview show in AI. He is known for preparing for weeks, reading the guest's papers, and asking the follow-up question most hosts skip. That is why top researchers and CEOs, from AI lab leaders to economists and historians, sit down with him for two or three hours. Episodes are dense with ideas about scaling, compute, AI timelines, alignment and what AI means for the economy, and each comes with a full transcript.

He is independent: the show is his own business, not part of a VC fund or media company, and he pushes back on guests. The Apple catalogue lists 141 episodes since 2020, so the pace is slower than a weekly news show, which is the price of the preparation.

The drawbacks are real. Listener reviews note he talks fast and jumps between topics, and the show assumes you already know the basics. Its 4.4/5 Apple rating is lower than most shows here, largely because of that pace.

Pick it if you want to understand where frontier AI is heading from the people steering it. Skip it if you are new to AI; start with the AI Daily Brief or Hard Fork.

Latent Space calls itself "the podcast by and for AI engineers", and it earns the title. Hosts swyx and Alessio Fanelli interview the people building models, coding agents, GPU infrastructure and AI products, and they ask practical questions: what broke, what it costs, what actually works in production. Guests have included people from OpenAI, Anthropic, Google, Meta, Modular's Chris Lattner and Answer.ai's Jeremy Howard.

It publishes weekly (226 episodes) and pairs each show with detailed notes on the Latent Space Substack. The team also runs the AI Engineer conference, and the show claims more than 10 million readers and listeners in 2025.

Two cautions. Alessio is a venture investor, so some guests are companies in or near his portfolio; the show is usually open about this, but it is not a critical outsider. And it is hard for newcomers: listeners note it assumes a lot of domain knowledge.

Pick it if you write code with AI or build AI products. Skip it if you want general AI news or big-picture debate.

TWIML (formerly This Week in Machine Learning & AI) is the veteran of AI podcasting. Sam Charrington has published 795 episodes since 2016, interviewing researchers and engineers about specific papers and projects: world models, voice agents, materials science, AI security and more. Each episode is usually about an hour, focused on one guest and one piece of work.

Charrington is good at asking a researcher to explain their method simply, then pressing on what it means in practice. That makes TWIML the best show for keeping up with research without reading every paper. It also has one of the strongest listener ratings on this list, 4.7 out of 5 from 425 ratings.

It is less exciting than the big-personality shows. There is little debate or news commentary, and some episodes are narrowly technical, which is great if the topic interests you and skippable if not.

Pick it if you work in ML or data science and want one well-explained research idea a week. Skip it if you want news, business strategy or entertainment.

If you only have time for one AI show a week, make it Last Week in AI. Andrey Kurenkov, who has an AI PhD from Stanford, and Jeremie Harris, who works on AI and national security, walk through the week's model launches, research papers, business news and policy in segments. A recent episode covered new models from Anthropic, Google and OpenAI, agents misbehaving during safety evaluations, and EU AI Act rules.

The two hosts complement each other: one explains the technical side, the other the safety and geopolitical side, and they cover papers most news shows ignore. It is independent and has a 4.6/5 rating from 349 listeners.

The format is long (often well over an hour) and moves fast through many stories, so depth on any single topic is limited. A few reviewers mention strong language, which matters if you listen at work.

Pick it if you want a complete, trustworthy weekly summary including research and policy. Skip it if you want deep single-guest interviews.

Nathan Labenz is one of the most curious hosts in AI. The Cognitive Revolution mixes interviews with AI builders and researchers with discussions of what AI means for society, from agent-run businesses and agent memory to US-China relations and frontier model safety evaluations. Labenz uses the tools heavily himself, so he asks informed, practical questions, and he takes AI risk seriously without being alarmist.

With 376 episodes, it has a large back catalogue, and newer episodes include regular news panels with commentators such as Zvi Mowshowitz. It is one of the few shows that connects technical progress to business and policy consequences in the same conversation.

The drawbacks: episodes are long (often close to two hours), the schedule is busy and uneven, and the show is part of the Turpentine network, which is linked to the venture world. The Apple rating base is small.

Pick it if you already follow AI and want thoughtful, wide-ranging conversation. Skip it if you need short episodes or beginner explanations.

Machine Learning Street Talk is the most academic show on this list. Tim Scarfe and regular co-host Keith Duggar hold long conversations with scientists about AI, cognitive science, neuroscience and philosophy of mind. Recent episodes covered tabular foundation models with Frank Hutter, the limits of prediction-based AI, robotics, and interpretability. The show's stated goal is to examine ideas "with the hype surgically removed".

That sceptical stance is its value. Where many shows cheer every new model, MLST asks whether it really understands, generalises or reasons, and gives airtime to critics of scaling alongside believers. The YouTube versions are well produced, with visuals and chapter markers.

It is hard going for newcomers: episodes often run close to two hours and assume you know the vocabulary. The philosophical tangents will not suit someone who only wants practical tips.

Pick it if you have a technical or research background and want careful, critical discussion. Skip it if you want news or practical how-to content.

The AI Daily Brief is the best way to build a daily AI news habit. Nathaniel Whittemore (NLW) publishes almost every day, and each episode takes one or two stories, such as a new model launch, a business trend or a policy fight, and explains why they matter. A recent episode compared Anthropic's and OpenAI's newest models on benchmarks, real work and model personality. More than 1,100 episodes since April 2023 make it one of the most consistent shows in AI.

It is easy to follow, which is why it has one of the best ratings here: 4.7 out of 5 from 855 listeners. Whittemore also runs an AI training business, and the show is ad-supported with frequent sponsor reads.

The limits: it is commentary from one person rather than expert interviews, so depth is limited, and its business-and-adoption lens means less on research or harms. Some reviewers want more attention to AI's wider impact.

Pick it if you want 20 minutes a day to stay current. Skip it if you want deep technical detail; pair it with Latent Space or TWIML.

Hard Fork has been the most accessible good tech podcast of the AI era. Kevin Roose of The New York Times and Casey Newton of Platformer made AI news fun without dumbing it down, mixing interviews with lab leaders, hands-on tests of new tools and plenty of jokes. It has by far the largest rating base on this list (5,507 Apple ratings at 4.3/5), and its journalists have no financial stake in the companies they cover.

That era just ended. Roose and Newton recorded their final episode, released on 17 September 2026, and are launching an independent AI show called Machine Gods, distributed with NPR, in the week of 19 October 2026. Newton says it will publish twice a week. The New York Times says Hard Fork will return with new hosts, but it has not said who.

We have lowered its reliability score until the new line-up proves itself. The 215-episode back catalogue remains a great introduction to how AI reshaped tech.

Pick it if you are new to AI and want an entertaining weekly show; follow Roose and Newton to Machine Gods if they were the draw. Skip it if you want technical depth.

Practical AI does what the name says: it focuses on making AI "practical, productive and accessible". Hosts Daniel Whitenack, a data scientist and founder, and Chris Benson, an AI strategist, talk with engineers and business leaders about real deployments, tools and trends, from LLM apps to getting found in AI search. Episodes are usually under an hour and the tone is friendly.

It has run weekly for 374 episodes and sits in the middle ground between beginner news shows and deep technical podcasts, which makes it a good fit for developers, analysts and managers who want usable ideas without a research background.

Its weakness is that it rarely surprises. Conversations are sensible rather than sharp, and it has fewer headline guests than Dwarkesh or Latent Space. The 4.3/5 rating is solid but not standout.

Pick it if you want a steady weekly show about applying AI at work. Skip it if you already follow the technical shows; you will hear little new.

Lex Fridman's podcast is the biggest show on this list by audience, with 12,211 Apple ratings at 4.7/5, and it hosts some of the most famous AI interviews ever recorded. Episodes often run three hours or more, which gives guests room to explain their thinking in full.

For an AI-focused listener, though, it is no longer the first choice. The show covers everything from psychiatry and history to sport and politics; recent episodes included a UFC fighter and a Civil War historian. Apple lists it as updating roughly monthly, so AI episodes are occasional.

Fridman's gentle interview style also cuts both ways. Guests open up, but he rarely challenges them, so powerful guests can make claims with little pushback. Its technical AI episodes, such as with programmers and researchers, remain excellent when they appear.

Pick it if you like very long-form conversations and will pick out the AI episodes. Skip it if you want regular, focused AI coverage; Dwarkesh does that better.

No Priors is the AI podcast for the startup world. Sarah Guo, founder of the AI-focused venture firm Conviction, and serial investor Elad Gil interview founders and researchers about where AI businesses are going: recent topics include diffusion models for faster inference, chip design, data infrastructure and brain-computer interfaces. Episodes are shorter than most interview shows (often under 45 minutes) and it publishes weekly.

Both hosts are sharp and well connected, so guests are strong and the business analysis is useful if you are building or funding an AI company.

Independence is the main issue. The hosts are active investors in AI, and some guests run companies they have backed or might back. That does not make the show wrong, but it means you should hear it as insiders talking, not neutral analysis.

Pick it if you are a founder, investor or startup employee who wants to know what top investors are watching. Skip it if you want independent or critical coverage of AI companies.