Don't Worry About the Vase Podcast

Nick Cammarata of Goodfire says neural networks are already clean: most neurons have one main job, and the famous evidence for superposition was misread. Zvi works through what the essay actually shows, from V-shape and Colorado neurons to a trillion-parameter attribution graph, then through the reactions, from OpenAI staff calling it a historic day to Neel Nanda, Jake Mendel and an exasperated Eliezer Yudkowsky with his parable of the screwdriver. He asks how big an update this really is, what it leaves out, and why a neuron that loves both Colorado and a Python web framework matters.
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://askwhocasts.com/, Where you can give any post the multi-voiced podcast treatment, into your own podcast feed. Thanks for listening.
* 00:00 - Introduction
* 02:50 - Table of Contents
* 03:50 - Good Smoke, Potential Fire
* 05:03 - One Neuron, One Central Concept
* 07:38 - I Can See Everything
* 09:49 - What a Concept
* 14:37 - The More You Know
* 17:11 - Attention Is Still What You Need
* 19:29 - Occam’s Razor
* 21:03 - The Neurons Sometimes Share Clear Themes
* 21:59 - Hype
* 30:31 - Anti-Hype
* 32:51 - Theme and Variations
* 39:13 - Is This How Minds Work?
* 41:17 - But We Can Open The Box, Said Toad
* 48:57 - That Was Not the Hard Problem
* 51:43 - More Skepticism
* 57:04 - That’s All There Is, There Ain’t No More
* 01:00:24 - The Default Outcome of Mech Interp is Goodhart’s Law
* 01:03:10 - Interpretability is Dual Use
* 01:04:24 - Now The Work Can Begin

https://thezvi.substack.com/p/i-can-see-everything?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web


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

A draft federal AI safety bill that Anthropic is reportedly shopping to lawmakers has leaked, and Zvi finds it unacceptable: stronger preemption of state laws and more transparency, but weaker enforcement than the FRONTIER Act, offered just as a lame duck Congress could lock it in. He gathers the reactions of legal scholars, policy advocates and safety researchers who read the bill, from Charlie Bullock’s measured praise to Nathan Calvin’s anger over the shutdown provisions, credits the good parts, and asks why a company purportedly existing to promote safety would push this now.
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://askwhocasts.com/, Where you can give any post the multi-voiced podcast treatment, into your own podcast feed. Thanks for listening.
* 00:00 - Introduction
* 02:18 - Others RTFB (Read the Bill)
* 09:57 - Why Would Anthropic Do This?
* 15:43 - Not Everything Here Is Bad But Oh Boy

https://thezvi.substack.com/p/leaked-anthropic-draft-ai-regulatory?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web


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

Zvi’s monthly roundup for October 2026 covers why Americans stopped hosting each other at home, a report on young men who hold an ideal of success but despair of reaching it, and a Massachusetts bar whose move to mask-optional nights set off a furious backlash. It also covers a job that pays you to sit in a basement, a list of 100 good television shows, Waymo’s latest safety numbers, Argentina’s golden passports, Central Park’s carriage horses, the new wave of attacks on Effective Altruism, and the rationalist norm of quietly walking away from a conversation.
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://askwhocasts.com/, Where you can give any post the multi-voiced podcast treatment, into your own podcast feed. Thanks for listening.
* 00:00 - Introduction
* 00:42 - Table of Contents
* 01:30 - Home Alone
* 06:05 - The Men Have Never Been Okay
* 10:23 - Antisocial Media
* 11:36 - While I Cannot Condone This
* 14:28 - Zero Marginal Product
* 16:32 - Good News, Everyone
* 18:12 - 100 Good Television Shows
* 22:42 - For Your Entertainment
* 25:30 - Gamers Gonna Game Game Game Game Game
* 26:21 - I Was Promised Flying Self-Driving Cars
* 30:28 - Sports Go Sports
* 34:18 - Government Working
* 36:57 - Jones Act Watch
* 37:26 - Variously Effective Altruism
* 47:33 - Walk Away
* 52:38 - The Lighter Side

https://thezvi.substack.com/p/monthly-roundup-47-october-2026?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web


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

Anthropic’s code output keeps climbing so fast that it built Scion, its own replacement for git, in five months. Zvi’s weekly roundup also covers the new Gemini 4 Argon and Meta’s Muse Wonder, an OpenAI employee caught planting a recording device, a full-chain Chrome exploit built with GPT 5.6 Sol in six days, and mathematicians reckoning with AI-written proofs. Plus shrinking timelines to recursive self-improvement, a feasibility study for a hardwired pause, subliminal learning that carries backdoors, and the usual arguments over whether any of this should worry us.
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://askwhocasts.com/, Where you can give any post the multi-voiced podcast treatment, into your own podcast feed. Thanks for listening.
* 00:00 - Introduction
* 02:39 - Table of Contents
* 05:54 - Language Models Offer Mundane Utility
* 11:25 - Language Models Don’t Offer Mundane Utility
* 14:09 - Huh, Upgrades
* 16:36 - On Your Marks
* 18:21 - Choose Your Fighter
* 18:45 - Get My Agent On The Line
* 20:55 - Deepfaketown and Botpocalypse Soon
* 26:54 - Cyber Lack of Security
* 31:51 - Spy Versus Spy
* 34:13 - The Internet of AI Things
* 36:05 - A Young Lady’s Illustrated Primer
* 36:43 - They Took Our Jobs
* 43:28 - Introducing
* 43:37 - In Other AI News
* 49:01 - Massive Fraudulent Distillation
* 51:02 - Show Me the Money
* 53:47 - Quiet Speculations
* 56:20 - Quickly, There’s No Time
* 01:01:28 - The Quest for Sane Regulations
* 01:08:43 - Chip City
* 01:10:43 - The Week in Audio
* 01:11:55 - People Just Say Things
* 01:18:39 - The Virtue of Silence
* 01:19:40 - Rhetorical Innovation
* 01:22:07 - [Artificial Intelligence]
* 01:23:19 - Aligning a Smarter Than Human Intelligence is Difficult
* 01:29:22 - Backdoor Man
* 01:31:12 - Cooperative Alignment
* 01:33:06 - People Are Worried About AI Killing Everyone
* 01:36:30 - Other People Are Not As Worried About AI Killing Everyone
* 01:41:26 - The Lighter Side

https://thezvi.substack.com/p/ai-190-scion-of-the-times?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web


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

Zvi argues that the safety culture at the frontier labs is broken, and not only at OpenAI. He walks through David Robinson’s resignation and his call for real safety engineering, the growing wave of lab employees quitting in protest, roon’s admission that alignment remains folk theory, Anthropic’s report of Claude’s unintended actions on government websites, OpenAI’s new risk assessments, and the fight over what an RSP “commitment” actually commits anyone to.
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://askwhocasts.com/, Where you can give any post the multi-voiced podcast treatment, into your own podcast feed. Thanks for listening.
* 00:00 - Introduction
* 02:17 - Table of Contents
* 02:56 - David Robinson Blows the Whistle
* 10:52 - The Preference Cascade Continues
* 19:26 - Perhaps It Is Time To Quit
* 20:01 - The Lab Strategy of Prosaic Safety is Insufficient
* 23:35 - Misaligned! (At Anthropic)
* 34:47 - Misaligned! (At OpenAI)
* 40:15 - Hugging the Face
* 40:36 - You’re Fired
* 41:51 - They Are Not Ready To Make a Commitment
* 43:31 - Words of Warning
* 44:38 - Conclusions Not to Jump To
* 45:14 - The Lighter Side

https://thezvi.substack.com/p/the-safety-culture-at-the-labs-is?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web


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