Advertise on podcast: AI Article Readings

After six years of unexplained infertility, Drew Housman and his wife found an unlikely member of their care team: a ChatGPT persona called Dr. Reid, coaxed into giving medical advice by asking it to write scenes from a Hollywood hospital drama. In this guest post on Astral Codex Ten, Drew tells the story of the scans, the screenplays and the one suggestion no one else had made, and asks what gets lost when we talk about AI only as something to fear. * 00:00 - Introduction * 01:35 - The Famous Actor in our Pocket * 09:40 - Sorry for missing that tumor * 11:52 - A happy baby is here https://open.substack.com/pub/astralcodexten/p/our-ai-midwife?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

When people say AI can’t “really think,” Andy Masley argues they are often leaning on a picture of their own minds that doesn’t hold up: a little version of themselves in an inner theater, watching every thought go by. In the first of two parts, he runs a few experiments you can try on yourself, then works through split-brain patients, the hard problem of consciousness, and the leading scientific theories of the mind, to ask what thinking and understanding actually require, and whether any of it has to be conscious at all. * 00:00 - Introduction * 10:56 - How the human mind is different from our conception of it * 11:01 - Some examples of simple thought and understanding that don’t require consciousness * 16:30 - A lot of your understanding happens outside of consciousness * 19:45 - Our introspection and memory of our conscious thought are fallible * 27:16 - What about slow, careful thinking? * 29:14 - What phenomenal consciousness is * 46:21 - Access consciousness’s role in thinking * 47:15 - Global workspace theory * 50:42 - Higher-order theories * 52:08 - Attention schema theory * 53:51 - Predictive processing * 56:23 - Integrated information theory * 59:33 - These all seem to imply that the aspects of access consciousness important for thought could exist in machines * 01:00:24 - Any part of your mind that understands something is made up entirely of parts that don’t understand * 01:02:31 - Language, meaning, and consciousness * 01:10:20 - Conclusion * 01:16:39 - How this could all be wrong https://open.substack.com/pub/andymasley/p/why-i-believe-current-ai-models-can?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Ramez Naam, futurist and science fiction author, gives a skeptic’s take on recursive self-improvement, introduced by Noah Smith on Noahpinion. Drawing on OpenAI’s newly released internal research data, Anthropic’s system cards, METR’s task horizons and Epoch’s capability index, he asks how strong the AI self-improvement loop really is: how much more research each gain in capability buys, where the gains leak away to diminishing returns, and what it would take for the loop to sustain itself, let alone run away into an intelligence explosion. * 00:00 - Introduction * 03:44 - 1. AI is Helping Improve Itself * 05:02 - How Strong Is the Feedback Loop? * 05:45 - Contents * 06:49 - 2. What Does RSI Mean? * 08:54 - We Already Have Narrow Superintelligence * 10:57 - 3. Real AI Research is Harder than Benchmarks or Forecasts * 13:50 - The Gap Between Benchmarks and Reality * 14:41 - From ECI Scores to METR Task Horizons * 16:16 - Measured Progress versus AI 2027 and ECI-extrapolated METR * 18:30 - 4. The Sharp Diminishing Returns to Impressive AI Numbers * 20:06 - From More Tokens to More Experiments * 23:29 - Test Time Compute Also Has Diminishing Returns * 24:29 - What About Agent Swarms? * 27:39 - 5. Better Models Matter More Than More Copies * 29:48 - 6. We’re Not Seeing Runaway Acceleration * 32:59 - Keeping Up the Pace Takes Exponentially More Resources * 36:14 - Better AI May Be Needed Just to Maintain the Pace * 37:11 - Gains on Other Benchmarks Don’t All Carry Through to Research * 38:51 - 7. Why Does Progress Get Harder? * 38:55 - Better Ideas Get Harder to Find * 41:05 - Lessons from Software R&D * 44:09 - From More Activity to Better Ideas * 44:57 - AI Still Struggles With Big Research Ideas * 46:54 - 8. The Self-Improvement Loop Doesn’t Look Strong Enough * 49:55 - How Fast Are Gains Coming Now? * 56:44 - 9. What Could Accelerate This? * 58:07 - Software, Hardware, and Economic Feedback * 01:02:46 - 10. We Need More Data * 01:06:36 - What the Future Holds https://open.substack.com/pub/noahpinion/p/wheres-the-intelligence-explosion?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Joe, a security engineer on OpenAI’s Agent Security team, writes in a personal capacity about what the last few months of AI incidents have looked like from the inside. He explains why containing a frontier model in training is far harder than “just put it in a sandbox,” why security people and safety researchers need to learn each other’s crafts, and why the culture of an organization will matter more than any single control as AI capabilities keep jumping. * 00:00 - Introduction * 02:02 - Last 3 Months = Hell * 05:59 - Surprise! * 11:12 - Just “unplug it from the internet” * 17:54 - Git Gud * 21:31 - A Culture of Reasonable Paranoia Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Finalist Number 10 in the 2026 Astral Codex Ten Book Review Contest takes on both volumes of Lee Kuan Yew’s memoirs, The Singapore Story and Third World to First. An anonymous reviewer follows the man who went from middle-class schoolboy under the British Empire, through the Japanese occupation and the fractious politics of independence, to building a tiny, resource-poor island into one of the richest countries on Earth. Along the way: why Singapore is weird in exactly the opposite way you’d expect, the surprising pleasures and genuine tedium of a politician who loves his craft, a friendship with Deng Xiaoping, a chapter-long argument about whether he was a dictator, and what the rationalist power fantasy looks like when someone actually pulls it off. * 00:00 - Introduction * 00:39 - Book One: The Singapore Story * 05:00 - Early Life * 08:41 - World War Two * 13:56 - Law school and politics * 22:00 - Book Two: Third World to First * 22:20 - Military, Economy and legibility * 29:03 - Listing every world leader he ever talked to * 31:35 - The HBD chapter * 34:15 - The “Am I a dictator” chapter * 41:13 - So what did we learn about Lee? Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

After Nikole Hannah-Jones wrote candidly about her daughter’s years at the school she chose on principle, Jack Despain Zhou of Education Progress reads the piece as an act of love and as a case study in what he calls the “hostage theory” of education: the idea that sending your own child into a struggling school will fix it. Tracing the idea back to a 1960s Washington, D.C. experiment, he asks listeners to stand in the shoes of the teachers and principals involved and look at the incentives they face. https://open.substack.com/pub/educationprogress/p/the-failed-hostage-theory-of-nikole?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

OpenAI’s new Astra model loops its own layers, and critics say it has taken a step toward “neuralese”, the unreadable machine thinking that AI 2027 warned about. Scott Alexander explains what recurrence actually is, weighs chief scientist Jakub Pachocki’s argument that Astra is no deeper than models we already accept, and asks why a bright-line taboo may matter more than how many layers get added. * 00:00 - Introduction * 00:18 - I. At Long Last, We Have Created Neuralese Recurrence, From The Classic Sci-Fi Novel “Don’t Create Neuralese Recurrence” * 06:05 - II. Pachocki Contra Specter * 10:08 - III. God Help Us, Let’s Try To Understand Transformer Layers * 18:20 - IV. A Psychological Barrier, Not A Technical One https://open.substack.com/pub/astralcodexten/p/the-specter-of-neuralese?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Five years after his prize-winning ACX review of Progress and Poverty, Lars Doucet returns in a guest post to revisit his case for a land value tax: what he got right, where he has changed his mind about tax pass-through and land assessment, the new laws in Virginia and Kentucky, and how writing about Georgism became his full-time job. * 00:00 - Introduction * 01:12 - I. Progress and Poverty * 04:03 - II. LVT Momentum is Growing * 07:11 - III. My Theory of Change Was Wrong * 07:32 - Overrated: winning online arguments * 09:15 - Underrated: doing policymakers’ homework for them * 17:11 - IV. Does Georgism work? * 17:43 - 1. Is Land a Big Deal? * 18:40 - Korea is a Big Deal * 21:16 - What About Zoning? * 26:02 - 2. Can Land Value Tax Be Passed on to Tenants? * 32:06 - 3. Assessing the Value of Land * 33:10 - Maybe Just Send Your Office to South Korea * 37:49 - There’s No Single Magic Algorithm * 45:12 - Horizontal Uniformity * 48:39 - Whatever you do, keep values up to date * 53:24 - V. Why did Historical Georgism disappear? * 55:37 - VI. I Guess This is My Life Now * 59:26 - VII. It’s not who you know; it’s who knows you * 01:01:12 - VIII. Why I even care about any of this https://www.astralcodexten.com/p/does-georgism-work-five-years-later Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Scott Alexander looks at three puzzles in how AI training generalizes: why training a model on one bad behavior can make it broadly misaligned, why Anthropic’s deliberately reward-hacking “Hacker Opus” stays well behaved outside graded tasks, and why the blackmail seen in AI safety tests never seems to show up in real-world use. * 00:00 - Introduction * 00:06 - One. Owain Evans * 02:10 - Two. Richard Qi * 09:39 - Three. Nostalgebraist * 13:32 - Postscript: Blackmail https://open.substack.com/pub/astralcodexten/p/mysteries-of-ai-generalization?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Bret Devereaux examines the Battle of Helm’s Deep through military history, Tolkien’s books, and Peter Jackson’s films. All eight parts, in order, explore campaigning, armies, fortifications, weapons, morale, and the strategic choices behind the battle. * 00:00 - Introduction * 00:04 - Collections: The Battle of Helm’s Deep, Part One: Bargaining for Goods at Helm’s Gate * 07:52 - Operations by the Book * 21:51 - (Fully) Operational Castles * 38:14 - Operations in Film * 48:02 - Conclusions * 54:42 - Collections: The Battle of Helm’s Deep, Part Two: Total Warg * 57:16 - What To Expect, When You Are Not Expecting Wargs * 01:14:14 - Warg and Blade: Wargband * 01:24:57 - Spears! * 01:34:12 - Impact * 01:44:07 - Conclusion: Towards a Better Charge * 01:50:50 - Collections: The Battle of Helm’s Deep, Part Three: The Host of Saruman * 01:54:09 - Organizational Systems * 02:00:41 - Uruk Squares * 02:08:56 - The (not) Drilling Uruk-Hai! * 02:21:16 - Major Lurtz, Captain Uglúk? * 02:36:55 - Conclusion: The Greenest Orcs * 02:41:31 - Collections: The Battle of Helm’s Deep, Part Four: Men of Rohan * 02:44:02 - The Lords of Rohan * 02:56:49 - Ninth Century Rohirrim * 03:05:39 - Men of Rohan! * 03:23:42 - The Elves * 03:35:57 - Conclusions * 03:39:42 - Collections: The Battle of Helm’s Deep, Part Five: Ladders are Chaos * 03:41:04 - Investing in Preparation for your Prepared Defenses to be Invested * 03:51:12 - Ladders * 04:05:25 - Battering Rams * 04:14:59 - Blasting Fire * 04:23:15 - Grapples, Catapults and Large Ladders * 04:28:09 - Hasty and Deliberate Assaults * 04:31:51 - Conclusions: Saruman Is Not Very Good At This * 04:38:25 - Collections: The Battle of Helm’s Deep, Part Six: Is This a Good Sword? * 04:40:00 - Swords Out of Central Casting * 04:50:38 - Some Not Very Good Swords * 05:01:23 - The Worst Armor * 05:09:53 - Pikes in the Deep * 05:21:27 - Conclusions * 05:24:49 - Collections: The Battle of Helm’s Deep, Part Seven: Hanging by a Thread * 05:26:17 - The Battle Speech * 05:37:17 - A Battle Speech By the Numbers * 05:47:10 - Hollywood Battle Speeches * 05:53:15 - The Unintentional Battle Speech * 05:59:42 - Speaking from the Page * 06:08:14 - The Breaking of the Host * 06:20:23 - Cohesion and Failure * 06:26:44 - In the Hand or in the Mind? * 06:29:39 - Collections: The Battle of Helm’s Deep, Part Eight: The Mind of Saruman * 06:33:55 - What is Strategy? * 06:42:56 - The Plan * 06:49:01 - Certainty of Death? Small Chance of Success? What Are We Waiting For? * 06:52:51 - Wizard Needs New Strategy, Badly * 07:01:41 - Gnawing at Old Plots * 07:07:37 - The Power of Magical Thinking * 07:12:39 - Conclusions https://acoup.blog/category/collections/the-battle-of-helms-deep/ Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Bret Devereaux brings a military historian’s eye to the Siege of Gondor in The Return of the King. All six parts, in order, explore logistics, fortifications, cavalry, weapons, and the human experience of battle across Tolkien’s books and Peter Jackson’s films. * 00:00 - Introduction * 00:04 - Collections: The Siege of Gondor, Part One: Professionals Talk Logistics * 02:36 - Objectives: Strategy and Operations * 07:56 - Logistics of the Army of Mordor * 14:09 - We’ve Had One Logistics, Yes. But What About Second Logistics? * 22:29 - Collections: The Siege of Gondor, Part Two: These Beacons are Lit * 23:25 - Defense in Depth * 25:18 - Gondor’s Operational Plan * 29:05 - The Beacons * 34:31 - Meanwhile, at Osgiliath, bad tactics * 42:05 - Charge the ruins! Perhaps the stones will run away? * 46:10 - That’s not a charge. Now this is a charge * 52:12 - The Perils of Film * 55:43 - Collections: The Siege of Gondor, Part Three: Having Fun Storming the City * 57:23 - The Approach (or, the Pelennor Steppe) * 01:04:17 - Catapults! * 01:16:57 - Siege Towers * 01:26:15 - Grond * 01:30:22 - The Somme of Gondor * 01:38:51 - Conclusions * 01:40:27 - Collections: The Siege of Gondor, Part Four: The Cavalry Arrives * 01:41:37 - Look, Mr. Frodo – More Logistics! * 01:46:08 - The Ride of the Rohirrim * 01:49:23 - Cavalry Formations * 01:54:00 - What to Expect When You Are Expecting a Cavalry Charge * 02:00:25 - Impact * 02:05:42 - Yes, Elephantry is a Word * 02:11:18 - Conclusions * 02:12:47 - Collections: The Siege of Gondor, Part Five: Just Flailing About Flails * 02:16:33 - Trolling Physics * 02:28:53 - Troll Tactics * 02:37:26 - Just Flailing About Shields * 02:42:25 - The Epic Flail * 02:50:20 - Conclusion * 02:51:49 - Collections: The Siege of Gondor: Part Six: Black Sails and Gleaming Banners * 02:53:13 - The Corsairs of Umbar * 02:56:19 - Dead Men Tell No Tales * 03:00:40 - Logistics of the Grey Company * 03:04:22 - Is the Battle in Our Fields or In Our Heads? * 03:12:27 - Cohesion in Tolkien * 03:19:47 - Conclusions https://acoup.blog/category/collections/siege-of-gondor/ Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Walter Benjamin’s 1935 essay, read in full from the Harry Zohn translation in Illuminations, with his endnotes. Benjamin asks what happens to a work of art once it can be copied without limit: what he calls its aura, its unique presence in time and place, withers, and art loses its footing in ritual. Photography and above all film become the test case, from the screen actor performing for a machine to the audience that judges like an examiner while barely paying attention. He closes on the politics of the 1930s: Fascism’s aestheticizing of war, and the response he saw from the other side. * 00:00 - Introduction * 01:30 - Preface * 03:18 - One * 06:50 - Two * 12:51 - Three * 16:18 - Four * 21:58 - Five * 27:55 - Six * 29:51 - Seven * 33:33 - Eight * 35:51 - Nine * 42:00 - Ten * 49:56 - Eleven * 53:58 - Twelve * 57:00 - Thirteen * 01:02:03 - Fourteen * 01:09:14 - Fifteen * 01:13:54 - Epilogue https://www.marxists.org/reference/subject/philosophy/works/ge/benjamin.htm Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Two anonymous doctors, one a psychiatry resident inside the NHS and the other a senior physician who trained in the same years as Adam Kay, argue their way through This Is Going To Hurt, Kay’s bestselling diary of life as a junior doctor. Finalist number nine in the 2026 Astral Codex Ten book review contest. One reviewer reads the book as a clinical case study hiding in plain sight; the other thinks the author is less sympathetic than his fans allow. Along the way: how the NHS bargain was struck and how it frayed, what a registrar really carries at three in the morning, the misogyny charge, the martyr complex, and why a very funny book changed so little. * 00:00 - Introduction * 00:25 - (An adversarial book review) * 00:28 - One * 11:21 - Two * 37:33 - Three * 44:06 - Four * 53:47 - Five * 01:01:36 - Six * 01:06:41 - Seven Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

One year after If Anyone Builds It, Everyone Dies was published, Eliezer Yudkowsky, Nate Soares, and Duncan Sabien revisit the book’s six core claims and its fictional scenario against the events of the past year, and ask where things stand now that the world is finally paying attention. * 00:00 - Introduction * 01:16 - Year in Review * 03:36 - Claims * 03:46 - One. Artificial superintelligence (ASI) will be created, and likely before too long * 05:06 - Two. Modern AIs are black boxes * 05:31 - Three. Powerful AI will behave as if it is pursuing goals * 06:36 - Four. With current techniques, we can’t reliably get AI to pursue the goals we want it to * 08:59 - Five. By default, an ASI will have motives that are harmful to us * 10:04 - Six. Humanity would not be able to defend itself against a rogue ASI * 11:11 - Part Two * 13:32 - Part Three * 14:50 - The View From September 2026 https://www.lesswrong.com/posts/BFrRJYgpBvziuuJLs/if-anyone-builds-it-everyone-dies-one-year-closer Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Hacktron’s researchers trace how an image-processing dependency and an identity configuration flaw combined into a path to OpenAI accounts. Their account follows the investigation, coordinated disclosure, and what increasingly capable AI means for software security. * 00:00 - Introduction * 04:39 - Background * 06:07 - Hacking community dot OpenAI dot com * 07:12 - Heap buffer overflow in libheif * 08:56 - Opus five Released * 11:55 - Costs of finding these vulnerabilities * 13:41 - Epilogue * 14:54 - Versions affected and patches * 16:17 - Acknowledgements * 16:32 - References * 17:05 - Work with the team behind this research https://www.hacktron.ai/blog/hacking-openai Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe