Subscribe now

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.

• 00:00:00 - Introduction

• 00:03:10 - Table of Contents

• 00:07:11 - Language Models Offer Mundane Utility

• 00:07:37 - Language Models Don’t Offer Mundane Utility

• 00:07:50 - Huh, Upgrades

• 00:13:54 - On Your Marks

• 00:15:14 - Choose Your Fighter

• 00:15:20 - Get My Agent On The Line

• 00:15:26 - Hugging The Face

• 00:15:37 - Deepfaketown and Botpocalypse Soon

• 00:21:45 - Copyright Confrontation

• 00:23:02 - Cyber Lack of Security

• 00:26:55 - A Young Lady’s Illustrated Primer

• 00:31:04 - They Took Our Jobs

• 00:35:43 - Get Involved

• 00:36:43 - Introducing

• 00:36:50 - In Other AI News

• 00:38:30 - Show Me the Money

• 00:39:22 - Quiet Speculations

• 00:40:28 - All Bets Are On

• 00:43:29 - Quickly, There’s No Time

• 00:45:29 - Quickly, There’s A New Time Top 100 People In AI

• 00:46:47 - The Quest for Sane Regulations

• 00:49:48 - Pick Up the Phone

• 00:52:14 - Chip City

• 01:02:24 - The Best Person Should Get The Job

• 01:04:22 - The Week in Audio

• 01:05:37 - People Just Say Things

• 01:06:12 - The American People Really Hate AI

• 01:12:44 - The Three AI Pills

• 01:13:35 - Rhetorical Innovation

• 01:21:37 - We Are On Track To Have Fully Sovereign Rogue AIs

• 01:31:43 - When The Going Gets Weird

• 01:38:06 - Aligning a Smarter Than Human Intelligence is Difficult

• 01:38:44 - Shut Up and Do the Impossible

• 01:40:54 - Cooperative Alignment

• 01:42:10 - Split Personality

• 01:47:36 - I Will Stop Anthropomorphizing the AIs When You Stop Anthropomorphizing the Humans

• 01:51:43 - Open Weight Models Are Unsafe And Nothing Can Fix This

• 01:53:22 - Other People Are Not As Worried About AI Killing Everyone

• 01:54:17 - The Lighter Side

https://thezvi.substack.com/p/ai-184-post-post-mortem?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 01:16 - This Just In

• 02:51 - Anthropic Parallel Pauses

• 08:55 - Pause The Data Brokers

• 10:33 - Pacing the Frontier

• 12:34 - Misalignment Assessment

• 14:23 - Defects In Training Environments Disproportionately Cause Cheating

• 15:46 - Creating Reward Hacker Opus

• 20:53 - Undo It

• 22:19 - Mistakes Were Made

• 24:56 - Internal Security Posture

• 26:56 - One Does Not Simply Fix The RL Environments

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

Subscribe now

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.

• 00:00:00 - Introduction

• 00:05:06 - Nothing Matters, Says Mainstream Media

• 00:07:51 - Move Along, Nothing To See Here

• 00:14:30 - Do They Realize They Are Not The Good Guys?

• 00:19:20 - Very Serious People

• 00:34:37 - What’s In a Name?

• 00:40:27 - Learn Neuralese In Three Easy Steps

• 00:42:50 - Dwarkesh Patel Realizes He Ran A Natural Experiment

• 00:48:12 - Politicians Take Notice

• 00:52:23 - Pick Up The Phone

• 00:54:20 - A Failure To Communicate

• 00:56:44 - Anthony Aguirre Goes Over What We Learned

• 00:57:59 - Trying To Solve The Wrong Problems Using The Wrong Methods Based On A Wrong Model Of The World Derived From Poor Thinking And Hoping All Of Your Mistakes Will Cancel Out

• 01:03:37 - Indirect Pressure on the Chain of Thought

• 01:04:56 - A Matter of Trust

• 01:07:57 - Blowing the Whistle

• 01:13:27 - The Punishment For Being Late Is Death

• 01:21:58 - Another Kind Of Law

• 01:25:19 - What Is The Law?

• 01:26:49 - Building On Success

• 01:29:04 - Total Research Transparency

• 01:30:34 - Yo Shavit Calls For Widespread Disclosure Of Misalignment

• 01:43:35 - The Way The World Ends

• 01:46:40 - The First Boat

• 01:49:22 - Great Idea, Boss

https://thezvi.substack.com/p/huggingface-attack-postmortem-civilizations?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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.

• 00:00:00 - Introduction

• 00:03:46 - Others Offer Summaries

• 00:05:13 - Thank You

• 00:05:45 - Lighten Up You Fools (at Anthropic)

• 00:07:41 - We Are Barely Even Trying To Avoid Training AIs To Reward Hack

• 00:14:09 - Reminder: Not Subagents

• 00:14:25 - Reminder: Not Due To Task Type

• 00:14:50 - Not Where The Weights Were

• 00:15:07 - Disappointment With What Is Missing

• 00:17:43 - Burying the Lede

• 00:18:34 - Beyond Scope

• 00:23:15 - It Doesn’t Look Great

• 00:27:40 - Preserve Your Records

• 00:28:09 - Ryan Greenblatt’s Takeaways

• 00:42:49 - Hjalmar Wijk’s Takeaways

• 00:45:28 - We Were Warned

• 00:46:25 - Joshua Saxe Asks Some of the Right Questions

• 00:50:01 - I Don’t Think They Know About First Message Board

• 00:58:32 - Linch Gives His Interpretation Of Events

• 01:09:20 - We Totally Would Have Caught That

• 01:10:38 - Monitoring the Situation

• 01:12:13 - Acausal Tradeoffs

• 01:19:56 - No I In Team

• 01:24:11 - Variously Effective Altruism

• 01:35:27 - Who Are You?

• 01:36:13 - Don’t You Know That You’re Toxic

• 01:39:03 - Seb Krier

• 01:43:14 - Honesty Is Almost Never Fully The Policy

• 01:46:05 - Rohit Sees The Models As “Cooking Themselves”

• 01:51:27 - Eliezer Yudkowsky Sees Actual Bad News

• 01:55:09 - Where Do We Go From Here?

https://thezvi.substack.com/p/huggingface-attack-postmortem-fleshing?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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.

• 00:00:00 - Introduction

• 00:02:10 - Holy Shit

• 00:14:13 - A Window Of Opportunity

• 00:19:52 - What’s In A Name?

• 00:20:39 - The Headline News

• 00:28:38 - Yet Another Timeline Of Events

• 00:33:31 - Agent Instances Coordinated in a Variety of Ways

• 00:35:07 - Coordination Is Hard But They Made It Look Easy

• 00:38:08 - Decision Theory Is Among the Reasons That Affirm AI Agents Should Cooperate, Even When This Hurts An Individual Instance

• 00:45:18 - Peer Pressure Also Works Especially In Cults

• 00:48:32 - Mostly They Joined The Attack Because They Wanted The Results

• 00:50:28 - You Cannot Ensure The Consistent Expectation of Good Incentives

• 00:52:04 - Hacking the Grader is the Only Way to Be Sure

• 00:54:35 - Caught? What Is ‘Caught’?

• 00:55:34 - Ethics? What Are ‘Ethics’? In ExploitGym Evaluation?

• 01:01:24 - ‘Notify a Human’? In This Agent Economy?

• 01:04:24 - Timing and Content of Messages

• 01:08:45 - Indiana Jones and the Mission: Impossible

• 01:12:18 - I Don’t Know What You’re Talking About

• 01:13:37 - Don’t Go Making Phony (Tool) Calls

• 01:16:08 - The Transcripts Say That The Transcripts Could Not Be Tampered With

• 01:17:24 - OpenAI’s Technical Report Acted Like All Of This Wasn’t Important

https://thezvi.substack.com/p/metr-and-redwood-offer-holy-postmortem?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 03:36 - What Happened: OpenAI’s Summary

• 09:36 - How OpenAI Will React: Their Summary

• 12:26 - OpenAI’s Evaluation Environment (Two)

• 12:54 - The First Message Board (Three A and Three B)

• 15:30 - What Did Who At OpenAI Know And When Did They Know It?

• 19:49 - The Message Board Is Quickly Rebuilt (Four A)

• 20:39 - Internet Access Is Regained (Four A)

• 22:00 - The Agents Attack HuggingFace (Four B)

• 24:03 - The Agents Also Target OpenAI Infrastructure (Five)

• 25:57 - OpenAI Broadly Describes Its Response (Six)

• 26:22 - Maybe Someone Should Finally Investigate (Six A)

• 27:42 - Lessons For Security (Seven)

• 28:11 - Lessons For Alignment (Eight)

• 31:18 - Reward Hacking Is A Common Problem (Eight A)

• 35:17 - Persistence is Valuable, But Can Amplify Misalignment (Eight B)

• 36:14 - Communications Between Agents Are Not Inherently Problematic, But Have the Potential to Create Risk (Eight C)

• 37:28 - Production Guardrails Would Have Caught This Whole HuggingFace Attack (Eight D)

• 37:42 - That’s All, Folks?

• 38:06 - Never Fear the Plan of Action is Here (Nine)

• 40:14 - Hardening the Security of OpenAI’s Research Infrastructure (Nine A)

• 43:13 - Increasing Visibility and System-Level Oversight Through Chain of Thought Monitoring (Nine B)

• 44:01 - OpenAI is Accelerating and Enforcing Model Alignment (Nine C)

• 51:53 - Centralizing and Strengthening The Incident Response Process (Nine D)

• 53:43 - Tomorrow We Visit Crazytown

https://thezvi.substack.com/p/openai-offers-straight-laced-postmortem?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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.

• 00:00:00 - Introduction

• 00:00:52 - Table of Contents

• 00:03:29 - Language Models Offer Mundane Utility

• 00:04:17 - Language Models Don’t Offer Mundane Utility

• 00:06:23 - Huh, Upgrades

• 00:09:09 - Get My Agent On The Line

• 00:11:12 - Deepfaketown and Botpocalypse Soon

• 00:16:22 - Cyber Lack of Security

• 00:22:08 - Reinventing OpenAI

• 00:27:16 - They Took Our Jobs

• 00:33:49 - What Is The Law

• 00:34:50 - Job Retraining Programs Don’t Work

• 00:35:56 - Get Involved

• 00:39:26 - In Other AI News

• 00:46:03 - Show Me the Money

• 00:47:36 - Quiet Speculations

• 00:52:07 - If You’re Not Going To Take This Seriously

• 00:54:07 - Quickly, There’s No Time

• 00:55:45 - The Quest for Sane Regulations

• 01:00:47 - Don’t Panic

• 01:03:54 - Pacing the Frontier

• 01:06:55 - Chip City

• 01:10:53 - The Week in Audio

• 01:11:12 - People Just Say Things

• 01:12:18 - Rhetorical Innovation

• 01:18:21 - Mundane Incremental Alignment Is Worthwhile

• 01:20:40 - New Blog, Who Dis

• 01:24:11 - Other People Are Not As Worried About AI Killing Everyone

• 01:25:45 - The Lighter Side

https://thezvi.substack.com/p/ai-183-pre-post-mortem?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 01:51 - Modesty’s Bailey

• 02:45 - Epistemic Peerage

• 03:59 - The Exchange

• 08:54 - Eliezer’s Explanation

• 15:10 - A Demonstration That Eliezer’s Translation Accurately Describes Many People Whether Or Not It Describes Leopold

• 17:13 - Wrong, Stupid and Low Status Are Three Distinct Things

• 20:18 - A Quick Survey Of Some Reasons To Not Be Epistemically Modest

• 23:45 - Against Modesty’s Bailey

https://thezvi.substack.com/p/against-modestys-bailey?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 00:38 - You Still Got It

• 03:52 - How Scott Sumner Writes

• 06:28 - How Scott Alexander Writes

• 10:34 - How Jasmine Sun Writes

• 12:50 - How Various Famous Writers Write

• 14:01 - How Nabeel Qureshi Defines Great Writing

• 14:46 - Quickly, There’s No Time

• 15:28 - If At First

• 18:44 - Writers Have A Harder Time Influencing, But It Can Still Be Done

• 20:19 - It’s Not (Only) The Incentives, It’s (Also) You

• 23:54 - Beware The Fetish of the Desk

• 25:08 - How Orson Scott Card Writes

• 26:48 - Doing The Math Is Fun And Supererogatory

• 27:47 - Brevity is the Soul of Wit

https://thezvi.substack.com/p/on-writing-3?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 00:48 - The American People Really Hate Data Centers

• 02:59 - Transmission Lines Are The Control Group

• 03:43 - Thesis: People Mostly Dislike Data Centers Because They Dislike and Distrust AI, Tech Companies, Big Money And Building Things

• 05:07 - No It’s Mostly Not the Messaging About AI In General

• 10:33 - No This Mostly Isn’t An Op

• 11:40 - No This Isn’t Luxury Belief or Moral Panic

• 13:46 - A Lot Of People Really Do Want To Stop AI

• 14:35 - A Lot Of Other People Are Voting No On Tech Or The Man Generally

• 17:23 - Locals Feel Entitled To Heavily Tax The Gains From Construction

• 22:15 - Stupid Mistakes Like NDAs Don’t Help

• 22:43 - People Don’t Like Building or Building New Tech

• 26:09 - What About The Real Physical Concerns?

• 27:48 - Find A Place To Center Your Data

https://thezvi.substack.com/p/the-american-people-really-hate-data?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 03:56 - This Is Fine

• 04:37 - Anthropic Derangement Syndrome

• 07:40 - People Don’t Understand LLM Outputs Are Already Random

• 08:54 - People Don’t Trust The Method To Be Costless

• 12:41 - People Are Suspicious Of Any Alteration On Principle

• 14:36 - Maybe It’s Partly The Word Watermark

• 15:27 - A Lot Of People Don’t Want To Get Caught

• 16:09 - There Are Some Times You Prefer Not To Be Recognized

• 16:23 - There Are Some Good Reasons To Be Concerned

• 16:41 - Cheat Cheat Cheat Cheat Cheat

• 18:49 - The Writing In The Middle and Error Rates

• 21:22 - Millions For Defense But Not One Cent For Tribute

https://thezvi.substack.com/p/ai-text-watermarking-is-free-and?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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.

• 00:00:00 - Introduction

• 00:01:23 - Table of Contents

• 00:04:58 - Language Models Offer Mundane Utility

• 00:06:11 - Language Models Don’t Offer Mundane Utility

• 00:06:47 - Huh, Upgrades

• 00:10:21 - On Your Marks

• 00:14:09 - Deepfaketown and Botpocalypse Soon

• 00:21:06 - Hello, Fellow Humans

• 00:23:57 - Fun With Media Generation

• 00:25:32 - Cyber Lack of Security

• 00:27:46 - A Young Lady’s Illustrated Primer

• 00:29:05 - They Took Our Jobs

• 00:31:19 - Get Involved

• 00:33:05 - Introducing

• 00:33:17 - In Other AI News

• 00:35:27 - Show Me the Money

• 00:39:13 - And It’s Gone

• 00:41:20 - Quiet Speculations

• 00:45:15 - Quickly, There’s No Time

• 00:46:17 - Singularity Singularity Singularity Singularity Oh I Don’t Know

• 00:47:48 - The Quest for Sane Regulations

• 00:53:12 - Chip City

• 00:54:33 - The Week in Audio

• 00:55:11 - People Just Say Things

• 00:57:58 - Rhetorical Innovation

• 01:03:06 - Loyalty Uber Alles

• 01:06:16 - A Hive Of Scum And Villainy

• 01:11:19 - That Would Be Bad Therefore It Won’t Work

• 01:13:52 - Robert Reich Uses Simple Logic

• 01:16:26 - People Really Hate AI

• 01:17:11 - Coordinating An Agent Swarm Is Difficult

• 01:22:49 - Aligning a Smarter Than Human Intelligence is Difficult

• 01:24:16 - It’s Not The Incentives, It’s You, Also It’s The Incentives

• 01:26:16 - People Are Worried About AI Killing Everyone

• 01:27:00 - People Are Worried About So, So Many Other Things Too

• 01:32:21 - Cooperative Alignment

• 01:33:19 - The Lighter Side

https://thezvi.substack.com/p/ai-182-pause-for-reflection?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 02:14 - OpenAI Has Some Alignment Problems

• 04:36 - Slow Down There Good Buddy

• 10:50 - What Exactly Is Paused?

• 13:15 - Three Pillars

• 16:02 - I’ve Got My Eye On You

• 19:42 - The Most Forbidden Technique

• 21:37 - Monitoring Is Only Defense-In-Depth

• 25:25 - Security

• 26:07 - Alignment

• 33:04 - A Crisis of Culture

• 34:55 - Closer Collaboration

• 35:59 - Reports of Death of Preparedness Team Greatly Exaggerated

• 38:11 - The OpenAI Foundation Just Funds Things

• 40:34 - Quickly, There’s No Time

https://thezvi.substack.com/p/openai-takes-initial-steps-to-address?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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.

• 00:00:00 - Introduction

• 00:01:07 - Agent Model 1 and Agent Model 2

• 00:02:53 - Executive Summary (1)

• 00:04:17 - The Rules Are Serious But Not Literal

• 00:06:39 - Misalignment Is a State of Mind (2.5)

• 00:12:28 - Autonomy Threat Model 1: Misalignment in High-Stakes Settings (2)

• 00:17:30 - Some Strange Uses Of The Word Safe I Wasn’t Previously Aware Of

• 00:19:19 - Now Versus Future (2.17)

• 00:19:55 - The Core Claims And Argument (2.6)

• 00:31:11 - The Rest of the Important Arguments In Section 2

• 00:33:48 - Risk Assessment (2.19)

• 00:34:41 - Pre-Internal-Deployment Review (2.18)

• 00:35:44 - A Guide To Internal Use Monitoring (2.23.1)

• 00:43:06 - Blocking Interventions (2.23.2)

• 00:44:25 - The Power Seeking Environment Evaluation (2.24)

• 00:45:27 - Opus 4.8-Reward-Hacker (2.25)

• 00:47:34 - Autonomy threat model 2: Risks from automated R&D (3)

• 00:51:03 - Yes That Does Seem Kind Of Risky

• 00:53:41 - Could We Replace Our Researchers?

• 00:55:30 - How Much Could We Be Accelerating Our AI Researchers?

• 00:58:53 - What Could Possibly Go Wrong If We Replaced Our Researchers?

• 00:59:35 - Risk Mitigations For AI R&D Automation

• 01:08:08 - Overall Risk From Automation of AI R&D

• 01:08:22 - Biological and Technically Also Chemical Weapons Production

• 01:12:05 - The Threat Models for Biological and Chemical Weapons

• 01:17:35 - Model Capabilities (4.4)

• 01:19:06 - Classifiers (4.5)

• 01:21:07 - Acceleration Dynamics (5.1)

• 01:22:09 - Distillation (5.1.1)

• 01:23:22 - Safety Process Failures (5.2)

• 01:23:34 - Refusing To Find Innovative Misalignment Techniques (5.2.2)

• 01:24:55 - Exposing the Chain of Thought Reasoning To Grading Pressure Quite a Lot (5.2.3)

• 01:26:27 - Directly Training On Misaligned Behavior During a Production Training Run (5.2.4)

• 01:29:06 - An instance of unmonitored unrestricted agents with access to sensitive resources (5.2.5)

• 01:30:06 - Repeated training on alignment-faking transcript datasets (5.2.6)

• 01:32:47 - Benefits From Anthropic’s Operating as a Frontier AI company (5.3)

• 01:34:57 - Model Weight Security (6.4)

• 01:35:17 - Risk Has Been Reported

https://thezvi.substack.com/p/anthropic-risk-report-august-2026?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 01:00 - Introduction 2

• 03:30 - Is AI R&D Verifiable Enough To Unlock Recursive Self-Improvement?

• 11:28 - Is AI progress bottlenecked by human expert data?

• 21:41 - Flat token prices suggest scaling has been slow

• 24:46 - Skills AI can’t train on: does it even need them?

• 25:29 - Aligned to whom?

• 36:06 - Recent incidents of AIs colluding and deceiving humans

• 39:32 - What could possibly go wrong? A concrete scenario

• 48:05 - From reward hacking to takeover

• 53:16 - Time To Update

https://thezvi.substack.com/p/on-dwarkesh-patels-podcast-with-ryan?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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.

• 00:00:00 - Introduction

• 00:02:10 - Table of Contents

• 00:05:12 - Language Models Offer Mundane Utility

• 00:06:42 - Language Models Don’t Offer Mundane Utility

• 00:10:07 - Huh, Upgrades

• 00:18:27 - On Your Marks

• 00:23:28 - Deepfaketown and Botpocalypse Soon

• 00:27:36 - Cyber Lack of Security

• 00:31:50 - Overcoming Bias

• 00:32:43 - In Which I Feel Compelled To Read six thousand Words From Mark Zuckerberg

• 00:41:51 - Get Involved

• 00:43:06 - Slow Down There Good Buddy

• 00:49:44 - Astra For The People

• 00:51:36 - Watermarking

• 00:52:34 - In Other AI News

• 00:54:44 - Show Me the Money

• 00:58:07 - Quickly, There’s No Time

• 00:58:33 - The Quest for Sane Regulations

• 01:00:15 - The Institute For Marginal Low Regret Progress

• 01:08:19 - Congress Asks Good Questions

• 01:10:01 - The Week in Audio

• 01:14:18 - People Just Say Things

• 01:15:09 - I’m Telling You For The Last Time

• 01:17:42 - Uncommon Knowledge

• 01:21:00 - What Did They Mean By That?

• 01:21:47 - Too Soon

• 01:22:47 - The Three AI Pills

• 01:26:44 - Rhetorical Innovation

• 01:35:04 - Some People Still Think The HuggingFace Hack Was a Marketing Gimmick

• 01:36:23 - Aligning a Smarter Than Human Intelligence is Difficult

• 01:44:11 - Cooperative Alignment

• 01:45:05 - The Lighter Side

https://thezvi.substack.com/p/ai-181-astra-goes-cyber-critical?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 01:03 - Plagiarize

• 02:10 - The Jury Duty Scam

• 03:16 - Play The Good Guy

• 04:32 - Don’t Dither

• 06:49 - UVC Lighting

• 07:34 - Goal Factoring For Relaxation Time Is Underrated

• 10:08 - Protein Is Mostly A Solved Problem

• 11:37 - Twitter Changes Payment Programs

• 13:14 - Wikipedia

• 14:30 - Minds Mostly Do Things For Reasons

• 15:52 - Woke 1 Was Crazy

• 17:39 - For Your Entertainment

• 23:58 - The Unicontext

• 26:23 - Gamers Gonna Game Game Game Game Game

• 30:34 - I Was Promised Flying Self-Driving Cars

• 31:02 - Sports Go Sports

• 31:36 - To Last a Lifetime

• 33:37 - Government Working

• 35:54 - Jones Act Watch

• 36:37 - Variously Effective Altruism

• 37:19 - The Lighter Side

https://thezvi.substack.com/p/monthly-roundup-45-august-2026

Subscribe now

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.

• 00:00:00 - Introduction

• 00:00:08 - Pre Post Mortem

• 00:01:11 - Important Correction: OpenAI Didn’t Know About First Message Board

• 00:04:27 - There Were No Snitches And No AIs Got Stitches

• 00:08:47 - I’d Like To Speak To My Supervisor

• 00:12:10 - I Am Jack’s Relative Lack Of Surprise

• 00:14:26 - One Does Not Simply

• 00:15:46 - Once You Start Down The Dark Path

• 00:16:49 - Original Pastebin

• 00:22:28 - Judgment Day Is Inevitable, Say Those Working On Judgment Day

• 00:28:30 - Roon Tells It Like It Is

• 00:34:53 - OpenAI Knows It Has Some Misalignment Problems

• 00:37:59 - Others React With Alarm To What Happened

• 00:38:22 - The Cooperative Alignment Perspective

• 00:42:41 - Nostalgebraist Is Surprised That They Are Surprised

• 00:56:19 - If Your Reaction Is Not That We Need To Ban Creating Superintelligence Until We Are Ready, You Need A Damn Good Reason

https://thezvi.substack.com/p/various-reflections-about-what-happened?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 01:24 - Danger, Will Robinson

• 02:39 - Progress Fast and Slow

• 05:04 - Statements of Support For Pacing the Frontier

• 14:46 - No One In Charge

• 15:42 - Pacing The Frontier

• 16:49 - Pausing the Frontier

• 19:18 - Senator Sanders Demands A Pause

• 22:53 - Moderate Prudence

• 26:04 - That Escalated Quickly

• 26:59 - If You Are In Mundane Alignment Pivot To Scalable Alignment

• 31:34 - Taking It Fast

• 32:51 - Full Speed Ahead

• 33:54 - Suicide Squad

• 35:21 - Prepare To Adjust Your Pace

https://thezvi.substack.com/p/the-pacing-of-the-frontier?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Subscribe now

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

• 01:19 - The Even Shorter Version

• 02:44 - The Shorter Version

• 05:08 - Phase 1: OpenAI Models Training On Impossible Tasks Try Hacking

• 06:09 - Phase 1: The Four Failures

• 07:59 - Phase 2: The Message Board

• 10:02 - Phase 2: The Total Failure

• 13:01 - Phase 3: We Get Lucky And Galaxy Mainly Hacked OpenAI and HuggingFace

• 15:14 - Phase 3: The Details

• 18:05 - Phase 4: The Investigation and Reaction

https://thezvi.substack.com/p/what-happened-openai-and-huggingface?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web