The recommendation engine in Apple Podcasts is bad at one specific job: finding shows where you come away knowing more than when you started. It's good at finding shows that are popular, or that resemble something you already played. That means three hosts reacting to the news, founder origin stories, and wellness shows that wear a lab coat to sell supplements.
Karpathy's HackerTube tried to fix the same thing for YouTube by ranking videos on whether thoughtful people actually recommended them, not on engagement. This is my attempt at the podcast version. I had a few AI research agents go through 24 "Ask HN: what podcasts do you listen to?" threads from 2011 to 2025 (about 1,400 comments), plus curated lists, faculty recommendations and Reddit. Then I deduplicated the results, scored them, and went through them by hand. 237 shows went in and 82 came out.
The filter was simple: the show should be researchers, engineers, historians or practitioners talking about work they actually understand. No hype cycles and no padding. A good show can still be entertaining, but anything built mainly to entertain is left out.
The list is split three ways:
- Core: subscribe. These are consistently good.
- Deep cuts: excellent but narrow, infrequent, or finished. For most of them, the archive is the best part.
- Pick episodes: don't subscribe. At their best these shows are top-tier, but the average episode isn't.
Every show links out to Apple Podcasts and Spotify, plus a starter episode or two where one clearly stands out.
Core
Generalist
Three academics, one topic, 45 minutes. 1,000+ episode archive across science, history and philosophy; the single best backlog in audio.
- Imaginary Numbers
- Pauli's Exclusion Principle
Tyler Cowen's hyper-prepared interviews. Highest ideas-per-minute of any interview show; guests range from physicists to novelists.
- Paul Graham
- Larissa MacFarquhar: Writing Altruism
Julia Galef's interviews on how we know what we know. Archive-grade; on hiatus.
BBC's statistics show. Short, rigorous takedowns of numbers in the news; trains numeracy.
Multi-hour, heavily researched interviews with researchers on AI, biosecurity, economics and policy. Full transcripts.
- Bryan Caplan: Stop reading the news
- Athena Aktipis on cancer, cooperation, and the apocalypse
Science & math
A working physicist's long interviews plus solo lectures. Guests span physics, biology, economics and philosophy.
- 63: Finding Gravity Within Quantum Mechanics (solo)
- Solo: The Crisis in Physics
Quanta's show: Strogatz and Levin with top researchers (Tao, Wilczek) on one open question per episode.
Jim Al-Khalili interviews eminent scientists about their work and careers. 360+ episodes.
Mind & life
Technical neuroscience-meets-AI conversations, 90+ minutes, with working researchers.
Careful interviews on cognitive science, animal minds and intelligence.
Computing & AI
The best-prepared interviewer in tech. AI researchers, historians, economists; the Reiner Pope and semis episodes are reference-grade.
- Carl Shulman on AI scale-up
- How an Anonymous Researcher Predicted AI's Trajectory
Engineering & systems
Story-driven interviews with builders (SQLite, SerenityOS, Zig). Rare releases, almost no misses.
Bryan Cantrill and co. on firmware, hardware bring-up and computing history, from people who did the work.
Jane Street's show: one hard engineering problem per episode (clock sync, networking, compilers). Zero marketing.
Richard Feldman on language and compiler design with Lattner, Kladov, Valim, TigerBeetle.
- José Valim on static types in Elixir
- TigerBeetle's Spectacular Jepsen Report, with Joran Greef
Oxide's earlier show: legends of hardware/software interface. Ended; every episode is worth it.
Working cryptographers on PQ crypto, lattices, real-world breaks. Most technical security show.
Economics & markets
Russ Roberts' 1,000+ book-length conversations on economics and epistemics. Weekly since 2006.
- Andrew McAfee on More from Less
- Nick Bostrom on Superintelligence
Bloomberg's practitioners explain how one specific corner of markets or the real economy actually works.
Patrick McKenzie (patio11) on the mechanisms of payments, banking, fraud and government systems.
David Beckworth with Fed officials and monetary economists. The monetary-policy plumbing show.
Corey Hoffstein's interviews with quant PMs. Highest-signal quant show; on hiatus, archive is gold.
Matt Levine and Katie Greifeld; precise securities law and market structure under the jokes.
History
Multi-hour deep dives (Blueprint for Armageddon, Supernova in the East). Most-recommended podcast on HN by a wide margin.
- Wrath of the Khans
- Twilight of the Aesir I & II (HH 69/70)
Mike Duncan's sourced, season-per-revolution narrative. The Russian Revolution season is the peak.
Mike Duncan's original: Rome start to finish. Ended; the template for narrative history podcasts.
Paul Cooper's documentary-length, source-driven episodes on how societies collapse.
- 20: Persia - An Empire in Ashes
- 16: Bagan Empire of Myanmar
Philosophy
Peter Adamson (LMU/King's) walks the whole history of philosophy, including Islamic and Indian traditions. 500+ episodes.
15-minute interviews with leading philosophers on one idea. The cleanest short-form philosophy.
Deep cuts
These are narrower. Pick the ones in your lane and skip the rest. If a show has ended, don't count that against it: Tides of History wrapped up in April 2026, and its archive is as good as it ever was.
Research psychologist interviews on reasoning, behavior change and evidence.
Long Now Foundation lectures by scientists and historians on long time horizons.
Interviews with authors of major new papers in Nature. Best way to track frontier results weekly.
Institute of Physics magazine show with working physicists.
A mathematician explains one favorite theorem per episode. Ended; ~95 episodes.
20-30 minute interviews with leading earth scientists.
2-3 hour whiteboard math with Fields Medalists (Borcherds, Freedman). On hiatus, completely evergreen.
SFI researchers on complex systems, scaling and intelligence. Dormant since late 2024 but evergreen.
Perimeter Institute's own theoretical physics interviews.
Journal-club depth on virology papers. 1,370+ episodes; dense.
Two professors interview top scientists about how discovery actually happens.
Biology professors on big conceptual questions in evolution and ecology.
Research-depth ML interviews (Chollet, Carlini). Strong editorial lean; long.
Nathan Lambert (Ai2 post-training lead) on post-training and open models.
Paper-by-paper interviews with alignment researchers. Full transcripts.
The only expert-level reinforcement learning interview show.
The practitioner's weekly security news show; best of the news-format shows.
Well-researched computing history from primary sources.
APL/J/BQN practitioners on array programming. Niche, HN-loved.
Deep interviews with database and language builders (DuckDB, incremental SQL).
Hosted by a Google TPU architect; academic and industry architects. Few episodes a year.
The best regular show on firmware, robotics and edge hardware.
Primary-source interviews on type theory and proof assistants. Very niche.
Dense weekly Postgres internals and operations.
Honest Rust adoption case studies (Rust for Linux, ClickHouse).
Zingales and McLean on where markets fail, with heavyweight guests.
PIIE's Chad Bown on trade economics; the tariff explainer.
Dean Curnutt with vol and rates practitioners on the price of risk. No stock picks.
Progress-studies interviews from the Works in Progress magazine team.
Trend following and systematic investing; the roundtables are practitioner-deep.
Holland and Sandbrook; multi-part series are substantive despite the banter.
Patrick Wyman's academically grounded history. Ended April 2026; successor show is Past Lives.
BBC's Apollo series built on mission audio and engineer interviews. Apollo 11 is season one of the feed, which is now titled "13 Minutes: Trouble with Hubble".
Picks up where History of Rome ends. Patient and thorough.
The history of the English language, word by word. Long-running and meticulous.
Narrative history of nuclear power; HN pick.
Philosopher Barry Lam builds real stories into philosophical arguments. Story-driven but rigorous.
Stanford's Robert Harrison in long literary and philosophical conversations.
Pick episodes, don't subscribe
These are the shows that show up all over recommendation threads and deserve to, but only for certain episodes. Search their archives by topic or guest instead of letting the feed decide what you hear.
Some excellent operator interviews; uneven.
- Naval Ravikant
- Gary Taubes on sugar
- Dacher Keltner on power
Deep clinical science episodes mixed with sponsor-heavy wellness. Pick episodes.
- Matthew Walker on sleep (parts I & II)
Top-tier when the guest is technical (Carmack #309, Jim Keller, Kernighan). Most recent episodes drift to politics/celebrity; pick episodes.
Large, strong archive of ML research interviews; recent run leans enterprise. Search by topic.
Long-running open source show; high HN mentions but uneven. Kernighan #484 is a highlight.
IEEE-backed, 600+ episodes; uneven. Search the archive by topic.
- Rich Hickey on Clojure
- Kent Beck on testing
- SE Radio 744: Danilo Krummrich on Rust in the Linux Kernel
Solid Python ecosystem interviews; practical rather than deep.
Huge archive, many vendor-pitch episodes. Search, don't subscribe.
- Federated Learning with Mike Lee Williams
- Slack data platform with Josh Wills
- Software Verification in the Age of Coding Agents
The NVIDIA, TSMC, Google and AWS series are top-tier tech history; recent topics drift.
Great investor episodes in the archive; recent run skews to founder/VC talk.
Strong when the guest is a researcher; frequent politics. Pick episodes.
- David Krakauer (Santa Fe Institute)
- #227 Knowing the Mind
- The Map of Misunderstanding
Caveats
This is based on popularity among a particular crowd (HN readers, academics, a few rationalist forums), with my editorial judgment on top of that. Chemistry, climate and genetics are thin because good shows on those subjects barely came up in the threads. If you know one that belongs here, tell me.
The list lives in a single JSON file that this page renders from, so corrections are welcome as PRs.