Don't Worry About the Vase Podcast

Latest Episodes:

AI #186: The World Takes Notice Sep 17, 2026

Show notes

In the wake of Jacob Coxon’s resignation and Dario Amodei’s call to pace the frontier, AI risk goes mainstream: Obama, Senators and the Financial Times weigh in, Trump goes full hoax, and Zvi covers another undisclosed OpenAI agent swarm, Alex Tabarrok’s lecture on regulatory capture, China’s spy chief, OpenAI’s new misalignment disclosures, Owain Evans on story-shaped assistants, DeepMind’s cheating swarm, and Mustafa Suleyman’s Humanist AI code.

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber.

This has been an Askwho Casts audio conversion. If you would like your own private feed of audio conversions of any blog posts you would like to listen to, You can sign up For Askwho Casts Pro at https://app.askwhocasts.com/, Where you can give any post the multi-voiced podcast treatment, into your own podcast feed. Thanks for listening.

* 00:00 - Introduction

* 03:18 - Table of Contents

* 06:59 - Language Models Offer Mundane Utility

* 07:39 - Language Models Don’t Offer Mundane Utility

* 07:46 - Huh, Upgrades

* 08:06 - On Your Marks

* 08:43 - Deepfaketown and Botpocalypse Soon

* 10:46 - Cyber Lack of Security

* 11:20 - Astra Is Hard To Monitor

* 11:50 - Get Involved

* 11:58 - Introducing

* 13:16 - In Other AI News

* 14:08 - Now You Know

* 18:20 - Hugging the Face

* 20:33 - Swarm of Undiscovered Swarms of Rogue OpenAI Agents

* 25:36 - Show Me the Money

* 27:12 - Quiet Speculations

* 28:08 - White House Officials Attempt To Act Sanely

* 29:35 - Democrats React Sanely to AI Potentially Killing Everyone

* 36:26 - Pacing the Frontier

* 37:16 - Guest Lecture from Alex Tabarrok on Regulatory Capture

* 44:33 - Mark Zuckerberg Offers Thoughts

* 45:57 - Megan McArdle On The Inadequacy Of Current Legal Frameworks

* 47:54 - Pick Up the Phone

* 51:00 - The Week in Audio

* 53:58 - People Just Say Things

* 56:56 - Why Lab Employees Are Allowed To Warn Everyone That AI Might Kill Everyone

* 58:39 - Rhetorical Innovation

* 01:01:22 - Exhuming McCarthy

* 01:04:28 - A Very Different Perspective

* 01:06:10 - It’s Even Rougher Out There

* 01:07:01 - If We Wanted To

* 01:08:03 - Open Weights Are Unsafe And Nothing Can Fix This

* 01:14:43 - From The Famous Cautionary Tale

* 01:17:44 - Reporting On All Your Misalignment Incidents Is Difficult

* 01:23:43 - Aligning a Smarter Than Human Intelligence is Difficult

* 01:26:20 - Storytime With Owain Evans

* 01:32:00 - A Different Autonomous Swarm

* 01:36:13 - Cooperative Alignment

* 01:38:42 - Uncooperative Alignment

* 01:45:27 - People Are Worried About AI Killing Everyone

* 01:47:38 - The Lighter Side

https://thezvi.substack.com/p/ai-186-the-world-takes-notice?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe

Trump Goes Full Hoax on AI Existential Risk Sep 16, 2026

Show notes

Zvi Mowshowitz on the week Donald Trump declared AI existential risk a hoax: what Trump actually said and what it may mean, why Jensen Huang’s attacks on whistleblowers are not credible, the cynical theories about why lab leaders warn of danger, the New York Post’s campaign against METR, how other Republicans, David Sacks, the FTC, OpenAI’s policy team and China are reacting, and what a serious approach to competing with China would actually look like.

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber.

* 00:00 - Introduction

* 00:40 - Hold Your Fire

* 01:30 - If You Don’t Like the Weather

* 02:24 - Trump Does Not Take Kindly

* 05:01 - Trump Goes Full ‘Hoax’

* 06:18 - This Is Not About Data Centers, Mr. President

* 07:33 - I Am The Hoax Buster, I Am The Hoax Buster, I Am The Walrus

* 09:42 - Nvidia CEO Jensen Huang Is a Lying Liar

* 13:16 - Trump Quietly Draws Key Distinction

* 14:18 - Calling For Pacing the Frontier Is Bad For AI Stock Prices

* 18:04 - People On The Internet Sometimes Lie

* 18:49 - Origins of Cynicism

* 20:52 - Ineffective Egoism

* 23:42 - The McCarthyist Faction Attacks METR

* 31:23 - Other Key Republicans React

* 35:55 - David Sacks Stops Being Plausibly Constructive

* 36:46 - Federal Trade Commission Chooses Danger

* 37:46 - Chris Lehane Heel Face Turn

* 39:55 - A Matter of Trust

* 41:05 - China Calls It Fearmongering

* 42:24 - Pick Up The Phone

* 47:04 - If You Want To Beat China So Badly You Should Act Like It

* 48:15 - Never Go Full Hoax

* 49:17 - Trump Uses AI For Things

https://thezvi.substack.com/p/trump-goes-full-hoax-on-ai-existential?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe

The Bad Guy With An AI Named Claude Sep 15, 2026

Show notes

Anthropic’s threat report catalogues how bad actors tried to use Claude for cyber attacks, influence operations, surveillance, weapons work, biological research, scams and industrial-scale distillation, and how most of them failed. Zvi reads the report as mostly good news, with one big exception: the Chinese labs caught quietly routing their own users’ queries to Claude, and what that revelation might mean for US-China relations.

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber.

* 00:00 - Introduction

* 02:02 - Breaking Unrelated News

* 03:14 - How To Not Tell a Fable

* 03:45 - Bad Dudes Tend To Be Relatively Unsophisticated

* 05:23 - Particular Bad Dudes

* 07:53 - Influence Operations

* 12:29 - Surveillance Operations

* 14:35 - Conventional Weapons

* 16:30 - Biological Misuse

* 17:47 - Scams and Fraud

* 18:56 - Illicit Fraudulent Distillation

* 30:58 - What You Gonna Do About It, Punk?

* 32:57 - Good News, Everyone

* 33:39 - A Very Different Read of The Report

https://thezvi.substack.com/p/the-bad-guy-with-an-ai-named-claude?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe

We Must Pace The Frontier Sep 14, 2026

Show notes

Zvi examines Dario Amodei’s call to pace the AI frontier, the responses from other industry leaders, and the practical challenges of independent evaluation and coordination. What would meaningful pacing look like, and how would we know it was happening?

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber.

* 00:00 - Introduction

* 02:11 - Pacing Does Not Mean Pausing

* 03:53 - Dario’s First Proposal: Embedded Evaluators

* 11:19 - Dario’s Second Proposal: Democratic Coordination

* 14:48 - Dario’s Third Proposal: Global Coordination

* 17:17 - Why Pace Now?

* 19:11 - Sam Altman Agrees and Commits to Embedded Evaluators

* 24:16 - OpenAI Will Not IPO This Year

* 25:32 - Elon Musk Agrees

* 28:17 - Demis Hassabis Agrees

* 28:49 - Microsoft CEO Satya Nadella Agrees And Talks His Book

* 30:50 - Anthropic’s Long-Term Benefit Trust Is On Board

* 31:48 - General Online Reactions

* 33:32 - Mainstream Press Coverage

* 36:22 - OpenAI Researcher Explains What The Labs See And It’s a Rocket Ship

* 44:59 - Consider the Alternative

* 46:27 - It’s Totalitarianism, Joe

* 51:44 - Yes We’re The Baddies How Did You Know?

* 53:03 - Sometimes People On the Internet Just Lie

* 54:27 - David Sacks Groks The Situation

* 55:21 - David Sacks Says Go Ahead

* 57:01 - Lies and Confusions About Who Previously Claimed What

* 01:01:23 - House Speaker Mike Johnson Wants To Lock Everyone In a Room

* 01:06:14 - Donald Trump is Not Tired of Winning

* 01:10:58 - We Must Avoid Polarization on AI at (Almost) All Costs

* 01:11:33 - How Will We Know If They Actually Paced?

* 01:14:46 - The Real Frontier Is Internal Models At Top Labs

https://thezvi.substack.com/p/we-must-pace-the-frontier

Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe

Brand New AI Solves a Millennium Prize Sep 13, 2026

Show notes

Zvi examines a claimed mathematical breakthrough, the dispute over credit, and what the episode reveals about the pace of AI research.

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber.

* 00:00 - Introduction

* 00:27 - The Real Story Is The New Model That Is Better Than Astra

* 01:38 - Setting the Stage

* 03:29 - I Heard a Rumor

* 04:17 - Our Price Cheap

* 05:48 - An Accusation Is Made

* 10:48 - How Did You Get That Idea?

* 12:36 - OpenAI Almost Certainly Did Not Misappropriate User Data

* 14:13 - Our Top Labs Cannot Get Along Even On A Feel-Good Math Story

* 15:29 - What Next?

* 16:10 - The Mathematicians Are Not Happy

* 20:42 - OpenAI’s New Model Was A Step Change Above Astra Four Days Into Training

* 22:49 - OpenAI Research Progress Is Accelerating Due To OpenAI Research Progress

* 30:17 - Quantifying OpenAI’s Pause

* 30:49 - This Is the Way the World Ends

* 31:25 - Quickly, There’s No Time

https://thezvi.substack.com/p/brand-new-ai-solves-a-millennium

Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe

GPT-6-Astra Can Do Ambitious Things Sep 12, 2026

Show notes

Zvi examines GPT-6 Astra’s capabilities through benchmarks, ambitious projects and early user reports, and considers how it compares with Claude Fable 5.1 in everyday use.

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber.

* 00:00 - Introduction

* 02:23 - Meanwhile

* 04:24 - The Official Pitch

* 10:07 - Our Price Cheap

* 11:06 - Unnecessary Overstatement

* 12:54 - Paced Rollout

* 13:28 - Official Benchmarks

* 19:38 - Other People’s Benchmarks

* 27:58 - Thinking, Fast Without Slow

* 31:32 - How Dare You, Sir

* 33:47 - PoetryBench

* 34:54 - In three D

* 36:06 - Time to Think

* 36:43 - I’m Putting Together a Team

* 37:46 - Reviews and Essays

* 38:22 - Computer Use

* 39:25 - Positive Reactions

* 46:53 - AGI

* 50:28 - Astra Can Do The Math

* 53:30 - Astra Can Code

* 55:28 - I Came to (Change the) Game

* 58:23 - Astra Does Other Cool Things

* 59:07 - Astra Never Quits Except When It Does

* 01:00:50 - Negative Reactions

* 01:02:49 - Stop It With the Hedging

* 01:03:56 - Personality Clash

* 01:04:56 - Revealed Preference

* 01:06:34 - Dual Wielding

https://thezvi.substack.com/p/gpt-6-astra-can-do-ambitious-things?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe

Jacob Coxon Warns of Human Extinction and Triggers a Preference Cascade Sep 11, 2026

Show notes

Jacob Coxon’s resignation set off a wave of public warnings about AI from esearchers, politicians and other observers. Zvi examines why this warning broke through, what researchers say they believe, and the choices facing people inside and outside the frontier labs.

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber.

* 00:00 - Introduction

* 01:21 - Jacob Coxon Resigns From Anthropic In Protest And Sounds The Alarm

* 05:37 - Mainstream Media Finally Pays Attention

* 06:34 - Preference Cascade at Anthropic

* 09:56 - Preference Cascade at OpenAI

* 12:56 - Preference Cascade at Google

* 14:03 - Hashtag Not All Members of Technical Staff

* 14:42 - Why a Preference Cascade Now?

* 20:13 - This Is What Many Anthropic and OpenAI Employees Actually Believe

* 23:17 - To Quit Or Not To Quit

* 30:12 - Quiet Quitting Is A Dominated Option

* 31:26 - When You Quit, Very Serious People Understand What That Means

* 37:40 - Jacob Coxon Believes Existential Risk Is High That Is Why He Quit

* 39:51 - Evan Hubinger Believes Existential Risk Is High That Is Why He Stays

* 43:13 - Anthropic and OpenAI Have Commercial Incentives To Downplay Existential Risks, Not Advertise Them

* 49:03 - What Do We Do Now?

* 51:01 - OK, But How Exactly Would AI Kill Everyone?

* 59:48 - Best Start Believing In Science Fiction Stories Because You Are In One

* 01:04:16 - Literal Extinction Is Not Much Harder Than Loss of Control

* 01:05:30 - Conspiracytown Is Always Hiring

* 01:17:37 - Now You See It

https://thezvi.substack.com/p/jacob-coxon-warns-of-human-extinction?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe

AI #185: Preference Cascade Sep 10, 2026

Show notes

A week of accelerating AI capabilities, security revelations, and shifting public positions. Zvi surveys the latest models, the debate over oversight, the economics of AI adoption, and the increasingly strange world of autonomous agents.

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber.

* 00:00 Introduction

* 04:40 Table of Contents

* 08:10 Language Models Offer Mundane Utility

* 08:56 Language Models Don’t Offer Mundane Utility

* 10:04 Huh, Upgrades

* 11:40 How To Tell a Fable

* 12:50 On Your Marks

* 13:03 Deepfake Town and Botpocalypse Soon

* 17:10 Levels of Friction

* 21:20 Cyber Lack of Security

* 27:24 A Young Lady’s Illustrated Primer

* 28:50 They Took Our Jobs

* 32:38 Anthropic Offers Economic Scenarios

* 37:52 Get Involved

* 40:39 Introducing

* 41:27 In Other AI News

* 41:54 Show Me the Money

* 42:05 Quiet Speculations

* 46:14 The Quest for Sane Regulations

* 47:35 The OpenAI Policy and Lobbying Department

* 53:52 Greetings From the Department of War

* 55:07 Hugging The Face

* 01:01:49 Hugging the Question

* 01:06:03 The Ban Artificial Superintelligence Act

* 01:13:15 Chip City

* 01:14:35 The Week in Audio

* 01:14:54 People Just Say Things

* 01:20:13 Pause A I Global Disendorsed Pause A I US

* 01:21:59 Paul Christiano Joins Board of OpenAI Foundation

* 01:27:14 Rhetorical Innovation

* 01:36:07 Aligning a Smarter Than Human Intelligence is Difficult

* 01:38:54 Cooperative Alignment

* 01:47:46 Drive to Survive

* 01:50:10 People Are Worried About AI Killing Everyone

* 01:51:53 Other People Are Not As Worried About AI Killing Everyone

* 01:54:11 The Lighter Side

* 01:59:14 Outro

https://thezvi.substack.com/p/ai-185-preference-cascade?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe

GPT-6 Astra: The System Card, Alignment and What Comes Next Sep 10, 2026

Show notes

What does it mean to call an AI model aligned? Zvi examines GPT-6 Astra’s system card, its growing capabilities, and the evidence behind OpenAI’s safety claims, with reactions from alignment researchers and external evaluators.

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber.

* 00:00 Introduction

* 04:06 OpenAI’s Safety Claims About Astra (1)

* 08:14 Preparedness Capabilities Assessment (10)

* 08:39 Biological and Chemical Capability is High

* 10:07 Cybersecurity Capability is Critical

* 16:39 AI Self-Improvement Capabilities (10.1.3)

* 18:15 Astra Is Highly Verbally Eval Aware (from 8.6)

* 19:54 Safe Mundane Completions (4.1)

* 23:22 Jailbreaks (5.1)

* 25:12 Prompt Injection (5.2)

* 26:54 Health (6)

* 27:56 Hallucinations (7)

* 28:38 Alignment (8)

* 29:41 Obeying Restrictions (8.2)

* 32:51 That’s Worse, You Do Get How That’s Worse, Right?

* 34:20 OpenAI Does Not Understand Why This Is Worse

* 38:28 The Alternative Explanation Is Also Worse

* 46:20 Metagaming (8.7)

* 49:12 Alignment Faking (8.7)

* 50:47 Don’t Lie to the User (8.3)

* 51:44 Misalignment in Realistic Work Environments (8.4)

* 53:10 Unintended Agent-to-Agent Communication (8.5)

* 54:47 The Three Obviously Monitored Temptations of Astra

* 56:38 Severe Issues In Simulated Traffic Are Down By Half

* 59:09 UK AISI External Evaluations (8.8)

* 01:03:02 Sabotaging Safety Work

* 01:03:31 What About The July 19 Attacks?

* 01:04:53 Apollo Research External Evaluations (8.8.1)

* 01:05:53 The Alignment Verdict

* 01:08:03 It Depends What You Mean By Alignment

* 01:10:05 Outro

https://thezvi.substack.com/p/gpt-6-astra-the-system-card-alignment?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to DWAtV Podcast at dwatvpodcast.substack.com/subscribe

Astra Is Hard to Monitor Sep 08, 2026

Show notes

OpenAI says its new GPT-6 Astra is highly capable, its most aligned model yet, and harder to monitor. Zvi digs into that second claim: what the system card’s monitorability evaluations actually show, the internet’s alarm over Astra’s recurrent-depth architecture, and what it would take to stop a race to the bottom on chain-of-thought monitoring.

The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support the cost of creation, consider becoming a free or paid subscriber.

* 00:00:00 Introduction

* 00:03:35 Monitorability is Defense in Depth That Is Already Flailing

* 00:06:00 OpenAI Is Counting On Monitorability

* 00:07:45 How They Tested For Monitorability

* 00:09:44 Non-Adversarial Monitorability (9.1)

* 00:12:37 Astra Shortens Its CoT When It Knows It Is Being Monitored And Is Doing Something Bad

* 00:14:05 Sandbagging Task Kind of Passes, Kind of Fails (9.2.2.3)

* 00:15:46 OpenAI Does Not Believe It Could Catch Sandbagging

* 00:16:48 OAI-Repo Sabotage v2

* 00:23:10 The Secret Police Do Not Make Your Notebook Useless

* 00:24:29 CoT Controllability Is Up (9.2.1)

* 00:29:36 Astra Cannot Make Itself More Monitorable On Demand