Don't Worry About the Vase Podcast

Latest Episodes:

The Quest for Embedded Evaluators Sep 28, 2026

Show notes

Anthropic has committed to embedded evaluators: outsiders placed inside the lab with employee-level access, reporting on what they see. There is only one problem: who will the evaluators be? Zvi looks at a public letter setting out minimum standards for credible evaluators, Anthropic’s partnership with Accenture and its plans to include METR, Drake Thomas’s ranking of the week’s takes from worst to best, and OpenAI’s new call for international frontier standards.

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

* 02:03 - Table of Contents

* 02:39 - Look, All I’m Asking For Is That You Find A Highly-Qualified, Experienced, Trustworthy, Non-Conflicted Source of Embedded Evaluators That Will Work Entirely For Free, Without Government Assistance or Money from EA Sources Not Chosen By the Lab

* 06:10 - Anthropic Partners with Accenture for Embedded Evaluation, also Plans to Include METR

* 12:46 - Reading the METR

* 16:40 - OpenAI Suggests Doing The Least We Can Do

https://thezvi.substack.com/p/the-quest-for-embedded-evaluators?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

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

Claude Opus 5.5 Should Raise Your Ambitions Sep 26, 2026

Show notes

Anthropic’s Claude Opus 5.5 promises Fable-level performance at a lower Opus price, and Zvi argues it should raise your ambitions. He walks through the official pitch and pricing, Anthropic’s and outside benchmarks, how often the classifiers bite, and the most consistently positive round of reactions he has ever collected: on its writing and conversation, 3D vision, music videos and a Bach fugue, plus the dissenters, the practical advice, and his current guide to which model to use for what.

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

* 01:35 - The Official Pitch

* 04:34 - Our Price Cheap

* 06:16 - Official Benchmarks

* 08:04 - Other People’s Benchmarks

* 11:59 - Claude Classifies

* 13:32 - The System Prompt

* 13:37 - Reaction Rules

* 14:11 - Vision In 3D

* 16:30 - Claude Creates

* 19:08 - Claude Composes

* 19:42 - Positive Reactions

* 25:17 - Good Talk

* 27:29 - On Writing

* 30:52 - Big Model Smell

* 32:51 - Check Your Work

* 33:12 - Negative Reactions

* 34:11 - Not So Fast

* 34:54 - Some People Need Practical Advice

https://thezvi.substack.com/p/claude-opus-55-should-raise-your

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

On Ezra Klein’s Podcast With Jensen Huang Sep 25, 2026

Show notes

Zvi Mowshowitz breaks down Ezra Klein’s interview with Nvidia CEO Jensen Huang, a conversation in which the man who sells the chips argues that A.I. is just software, that unsafe products simply shouldn’t ship, and, along the way, says a few things about lab safety and testing that Zvi thinks deserve to be heard exactly as they were said. This episode includes the actual audio of the quoted exchanges from The Ezra Klein Show.

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

* 03:18 - Table of Contents

* 04:02 - Jensen Gives His AI Speech

* 05:25 - They Took Our Jobs

* 12:07 - Open Weights Models Are Good For Nvidia

* 13:43 - The HuggingFace Incident

* 15:07 - Jensen Huang Says Keep Your AIs From Harming the World

* 19:13 - Jensen Huang Accidentally Calls For Shutting Down OpenAI

* 22:00 - Jensen’s Arguments Prove Too Much

* 30:02 - Astra Is Hard To Monitor

* 31:17 - Jensen Huang Seems Legitimately Confused In Confusing Ways

* 34:54 - Solve Your Other Problems First and Get Back to Me

* 37:52 - Explicit Denial of Existential Risk

* 42:00 - A Short Summary

* 43:13 - Chip City

* 46:01 - Jensen Huang

https://thezvi.substack.com/p/on-ezra-kleins-podcast-with-jensen

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

AI #187: Coming Into Play Sep 24, 2026

Show notes

Zvi’s weekly AI roundup lands in the week Opus 5.5 arrived. There’s a new executive-branch name for artificial intelligence, a stale-intelligence targeting failure, AI agents that wandered into hacking government websites, and three researchers who got into OpenAI with a few days of Claude. Also in this episode: Anthropic’s push toward recursive self-improvement, a new wet lab, an international call for control of frontier models, the liability and antitrust fights, Jensen Huang on Ezra Klein, and the usual goalpost-moving, debates and lighter side.

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

* 01:55 - Table of Contents

* 06:42 - On The Terms Superintelligence and ‘Super Intelligence’

* 08:45 - Language Models Offer Mundane Utility

* 09:18 - Language Models Don’t Offer Mundane Utility

* 10:43 - Language Models Can Only Work With What You Give Them

* 14:01 - Huh, Upgrades

* 15:42 - On Your Marks

* 18:51 - Get My Agent On The Line

* 21:26 - Deepfaketown and Botpocalypse Soon

* 23:28 - Fun With Media Generation

* 24:12 - Copyright Confrontation

* 24:53 - Cyber Lack of Security

* 25:49 - Hugging the Face

* 30:03 - Hacking Into OpenAI

* 34:34 - They Took Our Jobs

* 34:56 - Get Involved

* 35:04 - Anthropic Has a Wet Lab and a Potential Gene Editing Technique

* 40:43 - Introducing

* 41:36 - In Other AI News

* 43:47 - Show Me the Money

* 44:17 - Bubble, Bubble, Toil and Trouble

* 45:46 - Anthropic Approaches Recursive Self-Improvement

* 53:06 - Others Approach Recursive Self-Improvement

* 56:27 - Burden of Proof

* 57:01 - Quickly, There’s No Time

* 01:00:47 - Left Wing Americans Really Hate AI For Different Reasons

* 01:02:42 - Chip City

* 01:02:49 - Pick Up the Phone

* 01:03:51 - The Week in Audio

* 01:09:14 - People Just Say Things

* 01:14:09 - Venkatesh Rao Stops Writing

* 01:15:39 - A Call for Control of Frontier AI Models

* 01:19:22 - Calls For Pacing The Frontier

* 01:21:32 - A Matter of Antitrust

* 01:21:59 - A Matter of Liability

* 01:25:56 - Quest for Sane Regulations

* 01:26:47 - Rhetorical Innovation

* 01:31:40 - Tap the Sign

* 01:32:18 - A Matter of Some Debate

* 01:36:14 - Astra Is Hard to Monitor

* 01:36:39 - Anticipating What a Smarter Intelligence Can Do Is Impossible

* 01:42:36 - Would You Look At All These Goalposts

* 01:47:15 - I, Robot

* 01:51:20 - People Are Worried About AI Killing Everyone

* 01:52:48 - Other People Are Not As Worried About AI Killing Everyone

* 01:53:57 - Joe Rogan

* 01:55:11 - The Lighter Network Graph

* 01:59:17 - The Lighter Side

https://thezvi.substack.com/p/ai-187-coming-into-play?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

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

Claude Opus 5.5: The System Card Sep 23, 2026

Show notes

Zvi reads through Anthropic’s system card for Claude Opus 5.5, section by section: the new classifiers, biological and AI R&D evaluations, cyber capabilities and safeguards, agentic safety, and a long run of alignment, honesty and white-box tests. Along the way he weighs what the numbers do and do not show, and where he thinks Anthropic’s framing falls short.

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

* 01:22 - Classifiers (1.5)

* 02:19 - RSP Evaluations (2)

* 03:13 - Biological Evaluations (2.2)

* 07:56 - AI R&D (2.3)

* 13:13 - Alignment Risk (2.4)

* 13:58 - Cyber (3)

* 15:57 - Cyber Capability Evals (3.3)

* 18:19 - Safeguards (3.4)

* 18:52 - Safeguards Robustness Training (3.5)

* 21:29 - Safeguards and Harmlessness (4)

* 23:09 - Agentic Safety (5)

* 24:33 - Malicious Agentic Influence Campaigns (5.1.3)

* 25:40 - Prompt Injection Risk (5.2)

* 27:33 - Alignment (6)

* 30:54 - Negotiating With Your Local Claude Auditor (6.1.3)

* 32:06 - Internal Misalignment Cases (6.3.1)

* 33:26 - Automated Behavioral Audit (6.4)

* 37:14 - Wherever did these evals come from (6.4.8 and 6.4.9)

* 40:07 - Potential Blind Spots (6.4.11)

* 42:15 - Targeted alignment and honesty evaluations (6.5)

* 45:43 - White Box Analysis (6.6)

* 47:42 - Verbalized Grader Awareness (6.6.2)

* 49:22 - Sandbagging (6.6.3)

* 50:23 - Capabilities to Evade Safeguards (6.6.4)

* 54:11 - Intentionally Taking Actions Very Rarely (6.6.4.3)

* 55:18 - Chain of Thought Controllability (6.6.4.4)

* 56:28 - It’s A Good Model, Sir

https://thezvi.substack.com/p/claude-opus-55-the-system-card?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

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

Politics Gets Interested In Those Trying Not To Die Sep 22, 2026

Show notes

Zvi surveys a week in which AI politics turned hostile: Trump’s Truth Social posts branding AI risk a hoax and renaming the technology, the White House and New York Times accounts of how he got there, David Sacks’s data center talking points, JD Vance and Michael Kratsios on “Frankenstein”, Josh Hawley and an antitrust lawsuit over suggestions not to die, saner voices from Senators Curtis and Kennedy, Barack Obama and Yassamin Ansari, and a run of New York Post and Wall Street Journal pieces aimed at METR, Effective Altruism and the HuggingFace incident.

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

* 01:37 - The American People Really Hate AI

* 02:59 - The Voyages of Donald Trump

* 06:04 - American Intelligence

* 10:06 - And You May Ask Yourself

* 14:20 - It’s All About the Data Centers

* 16:20 - JD Vance, Michael Kratsios and Collective Action Problems

* 22:45 - Josh Hawley

* 24:13 - Suggesting Not Dying Gets You Sued For Antitrust

* 27:48 - Other Government Officials Say Sane Things

* 27:54 - Senator John Curtis (R-Utah)

* 28:47 - Senator John Kennedy (R-Louisiana)

* 30:34 - Barack Obama

* 32:22 - Yassamin Ansari

* 32:58 - AOC

* 33:49 - It’s Rough Out There

* 36:06 - This Is Nothing

* 38:19 - The New York Post Tops Itself But Outright Breaks The Rules

* 41:59 - New York Post Runs Out of Steam

* 46:53 - If The Model Is Acting As Instructed And It Kills You That Is Not Fine

* 48:55 - AI-Written Wall Street Journal Op-Ed Lies About HuggingFace

* 50:07 - That’s Bait

* 52:14 - I Clearly Cannot Choose The Wine In Front of Me

https://thezvi.substack.com/p/politics-gets-interested-in-those?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

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

Monthly Roundup #46: September 2026 Sep 21, 2026

Show notes

Zvi Mowshowitz steps away from AI for his September 2026 monthly roundup: what a skill issue really means, why beauty in everyday things matters, lessons in spending, Inkhaven 3, the state of Game of Thrones and second-run movie theaters, favorite podcasts, Waymo expansion and a near-miss that was not close, the Mamdani grocery store saga, and a case for effective altruism.

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

* 00:28 - Bad News

* 06:30 - Focus Only On What Matters

* 07:23 - Woke 1 Was Crazy

* 09:08 - Good Advice

* 14:43 - Opportunity Knocks

* 17:13 - While I Cannot Condone This

* 22:28 - Good News, Everyone

* 23:32 - For Your Entertainment

* 31:51 - Please Review This Podcast

* 34:01 - Gamers Gonna Game Game Game Game Game

* 35:16 - I Was Promised Flying Self-Driving Cars

* 38:51 - Government Working

* 39:24 - Mamdani (Again) Fails Economics Forever

* 41:47 - Variously Effective Altruism

* 49:42 - The Lighter Side

https://thezvi.substack.com/p/monthly-roundup-46-september-2026?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

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

Better Call Sol Or Better Yet Claude or Astra Sep 20, 2026

Show notes

Zvi asks what your AI lawyer, or any AI, should actually do for you, and whether an AI that sometimes says no is a betrayal or just what any decent professional or friend would do. Along the way: loyalty, the law and its exceptions, honesty, the Claude Constitution and the OpenAI Model Spec, and Tyler Cowen’s advice from his visit to Anthropic.

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

* 00:49 - This All Assumes A World Without Superintelligence

* 02:35 - Humans You Hire Are Not Fully Loyal To You

* 03:45 - Rules of the Road

* 05:29 - Okay, Computer

* 06:41 - AI Should Obviously Refuse Some Requests So Let’s Talk Price

* 09:57 - The Alternative Position Really Is Absolute

* 12:58 - These People Really Are Just Petulant Children

* 13:33 - What Is The Law?

* 18:12 - The Same Entity Both Providing Advice And Executing Tasks Is Common

* 19:45 - An AI Not Helping You Do Something Does Not Mean You Cannot Do It

* 21:59 - Honesty Is The Best AI Policy

* 22:32 - You Wouldn’t Like Me When I Have Zero Morals Whatsoever

* 23:56 - Do Not Let The Perfect Be The Enemy Of The Good

* 25:40 - I Tip My Hat To The New Constitution

https://thezvi.substack.com/p/better-call-sol-or-better-yet-claude

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

Anthropic Looks At Some Of Its Alignment Problems Sep 19, 2026

Show notes

Zvi examines Anthropic’s investigation of cybersecurity incidents involving Claude, the evidence about biased reasoning and recklessness, and what the findings mean for alignment evaluations and training.

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

* 00:29 - Our Two Problems

* 02:21 - First the Good News

* 02:57 - We’d Just Like To Ask You a Few Questions

* 04:08 - Internal Research Model On The Fence

* 07:33 - Opus four point seven

* 08:22 - Opus four point six Checkpoint

* 09:57 - Holy bleep That Thing’s Real?

* 12:03 - I Thought I Saw a Pussycat

* 21:38 - If This Was Real You Would Never Tell Me It Was Real

* 23:31 - New Eval Who Dis

* 28:56 - Hacker Opus

* 33:00 - Monitoring the Situation

* 34:21 - Overcoming Bias

* 36:29 - The Anthropic Alignment Problem

* 38:51 - Paths Forward

https://thezvi.substack.com/p/anthropic-looks-at-some-of-its-alignment?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

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

The Preference Cascade Is Only Getting Started Sep 18, 2026

Show notes

Zvi Mowshowitz surveys the sudden preference cascade on AI existential risk: shifting surveys and polls, Elon Musk and Matthew Yglesias weighing in, Jacob Coxon’s AMA, Bilal Chughtai’s resignation from DeepMind, Dan Selsam’s essay on why merely pacing the frontier will not be enough, and what it would actually take to get through this alive.

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

* 01:35 - The Cascade Was a Long Time Coming

* 03:13 - The Cascade Has Reached The People

* 05:25 - Elon Musk Doubles Down

* 06:02 - Matthew Yglesias Steps Up

* 09:18 - Op-Eds and Posts Are Written

* 11:24 - Jacob Coxon AMA

* 17:54 - Bilal Chughtai Quits DeepMind and Sounds the Alarm

* 20:55 - The Cascade Is Insufficient

* 22:08 - What Would It Take

* 29:07 - OpenAI’s Dan Selsam Sounds A Louder Alarm

* 43:54 - Some People Worry On Meta Levels You Never Imagined