Don't Worry About the Vase Podcast

Podcast episode for DeepSeek v3.2 Is Okay And Cheap But Slow. * 00:00 - Introduction * 00:44 - A Brief History of DeepSeek * 03:52 - Once More, With Feeling * 07:31 - Reading The Paper * 09:27 - Open Language Model Offers Mundane Utility * 12:25 - Those Benchmarks * 17:21 - Open Language Model Doesn’t Offer Mundane Utility * 18:45 - Open Language Model Does Do The Math * 19:59 - I’ll Get You Next...

Podcast episode for AI #145: You’ve Got Soul. * 00:00:00 - Introduction * 00:01:49 - Table of Contents * 00:06:17 - Language Models Offer Mundane Utility * 00:07:17 - Language Models Don’t Offer Mundane Utility * 00:08:31 - On Your Marks * 00:10:15 - Get My Agent On The Line * 00:10:49 - Advertising Is Coming * 00:12:11 - Deepfaketown and Botpocalypse Soon * 00:18:29 - Fun With Media Generation *...

Podcast episode for On Dwarkesh Patel’s Second Interview With Ilya Sutskever. * 00:00 - Introduction * 01:47 - Table of Contents * 02:19 - Explaining Model Jaggedness * 03:58 - Emotions and value functions * 05:36 - What are we scaling? * 06:47 - Why humans generalize better than models * 08:10 - Straight-shooting superintelligence * 09:57 - SSI’s model will learn from deployment * 10:59 - Alignme...

Podcast episode for Reward Mismatches in RL Cause Emergent Misalignment. * 00:00 - Introduction * 02:53 - Abstract Of The Paper * 04:15 - The Problem Statement * 06:16 - The Inoculation Solution * 08:48 - Cleaning The Data Versus Cleaning The Environments * 10:31 - No All Of This Does Not Solve Our Most Important Problems * 15:46 - It Does Help On Important Short Term Problems The Don’t Worry Abou...

Podcast episode for Claude Opus 4.5 Is The Best Model Available. 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. https://open.substack.com/pub/thezvi/p/claude-opus-45-is-the-best-model?r=67y1h&utm_campaign=post&utm_medium=web&showWelcomeOnShare=false Get full access to DWAtV...

Podcast episode for Claude Opus 4.5: Model Card, Alignment and Safety. * 00:00:00 - Introduction * 00:01:50 - Claude Opus 4.5 Basic Facts * 00:03:26 - Claude Opus 4.5 Is The Best Model For Many But Not All Use Cases * 00:06:02 - Misaligned? * 00:09:39 - Section 3: Safeguards and Harmlessness * 00:11:46 - Section 4: Honesty * 00:13:27 - 5: Agentic Safety * 00:21:01 - Section 6: Alignment Overview *...

Podcast episode for AI #144: Thanks For the Models. * 00:00:00 - Introduction * 00:02:24 - Table of Contents * 00:05:21 - Language Models Offer Mundane Utility * 00:05:50 - Language Models Don’t Offer Mundane Utility * 00:06:17 - Huh, Upgrades * 00:09:02 - On Your Marks * 00:10:53 - Choose Your Fighter * 00:11:21 - Deepfaketown and Botpocalypse Soon * 00:19:03 - What Is Slop? How Do You Define Slo...

Podcast episode for The Big Nonprofits Post 2025. * 00:00:00 - Introduction * 00:01:46 - Table of Contents * 00:08:23 - A Word of Warning * 00:09:40 - A Note To Charities * 00:10:44 - Use Your Personal Theory of Impact * 00:12:34 - Use Your Local Knowledge * 00:13:39 - Unconditional Grants to Worthy Individuals Are Great * 00:16:11 - Do Not Think Only On the Margin, and Also Use Decision Theory *...

Podcast episdoe for ChatGPT 5.1 Codex Max. * 00:00 - Introduction * 01:09 - The Famous METR Graph * 03:07 - The System Card * 04:04 - Basic Disallowed Content * 05:07 - Sandbox * 06:25 - Mitigations For Harmful Tasks and Prompt Injections * 07:06 - Preparedness Framework * 07:28 - Biological and Chemical * 08:40 - Cybersecurity * 14:24 - AI Self-Improvement * 17:17 - Reactions The Don’t Worry Abou...

Podcast episode for Gemini 3 Pro Is a Vast Intelligence With No Spine. * 00:00:00 - Introduction * 00:01:53 - There Is A Catch * 00:03:35 - Andrej Karpathy Cautions Us * 00:04:59 - On Your Marks * 00:17:01 - Defying Gravity * 00:18:32 - The Efficient Market Hypothesis Is False * 00:21:12 - The Product Of A Deranged Imagination * 00:25:44 - Google Employee Hype * 00:29:24 - Matt Shumer Is A Big Fan...

Podcast episode for Gemini 3: Model Card and Safety Framework Report. * 00:00 - Introduction * 01:31 - Gemini 3 Facts * 02:38 - On Your Marks * 03:54 - Safety Third * 06:05 - Frontier Safety Framework * 06:57 - CBRN * 10:13 - Cybersecurity * 11:51 - Manipulation * 17:42 - Machine Learning R&D * 20:14 - Misalignment * 23:07 - Chain of Thought Legibility * 23:26 - Safety Mitigations * 26:02 - They C...

Podcast episode for AI #143: Everything, Everywhere, All At Once. * 00:00:00 - Introduction * 00:01:47 - Table of Contents * 00:06:02 - Language Models Offer Mundane Utility * 00:06:58 - Tool, Mind and Weapon * 00:11:16 - Choose Your Fighter * 00:11:39 - Language Models Don’t Offer Mundane Utility * 00:16:07 - First Things First * 00:16:44 - Grok 4.1 * 00:20:37 - Misaligned? * 00:24:10 - Codex Of...

Podcast episode for Monthly Roundup #36: November 2025. * 00:00:00 - Introduction * 00:00:17 - Good News, Everyone * 00:01:34 - Good Advice * 00:02:28 - Where’s The Party? * 00:02:47 - Antisocial Media * 00:06:59 - Government Working * 00:18:24 - Jones Act Watch * 00:19:05 - Variously Effective Altruism * 00:22:32 - Great Taste, Less Filling * 00:25:45 - How You Do Anything * 00:26:42 - Bad News *...

Podcast episode for GPT 5.1 Follows Custom Instructions and Glazes. * 00:00 - Introduction * 00:23 - Their Announcement * 04:24 - Their Pitch on GPT-5.1 Instant * 06:26 - Their Pitch on GPT-5.1 Thinking * 09:38 - Now With Extra Glaze * 14:25 - Genuine People Personalities * 15:30 - The End Of The Em-Dash? * 17:35 - Turning A Dial And Looking Back At The Audience * 18:04 - System Card * 20:01 - On...

Podcast episode for On Writing #2. * 00:00 - Introduction * 00:27 - Table of Contents * 01:08 - How I Use AI For Writing These Days * 04:37 - Influencing Influence * 06:29 - Size Matters * 07:15 - Time To Write A Shorter One * 08:43 - A Useful Tool * 08:59 - A Maligned Tool * 11:08 - Neglected Topics * 13:35 - The Humanities Don’t Seem Relevant To Writing About Future Humanity? * 14:51 - Writing E...

Podcast episode for AI Craziness: Additional Suicide Lawsuits and The Fate of GPT-4o. * 00:00 - Introduction * 00:47 - The Latest Cases Look Quite Bad For OpenAI * 02:35 - Routing Sensitive Messages Is A Dominated Defensive Strategy * 03:36 - Some 4o Users Get Rather Attached To The Model * 05:10 - A Theory Of How All This Works * 07:53 - Maybe This Is Net Good In Spite Of Everything? * 11:19 - Co...

Podcast episode for AI #142: Common Ground. * 00:00:00 - Introduction * 00:01:34 - Table of Contents * 00:04:25 - Language Models Offer Mundane Utility * 00:07:32 - Language Models Don’t Offer Mundane Utility * 00:08:53 - Huh, Upgrades * 00:09:26 - On Your Marks * 00:09:55 - Copyright Confrontation * 00:11:25 - Deepfaketown and Botpocalypse Soon * 00:13:40 - Fun With Media Generation * 00:16:25 -...

Podcast episode for The Pope Offers Wisdom * 00:00 - Introduction * 00:22 - Wisdom Is Offered * 03:04 - The Context of The Meme Andreessen Used (If You Don’t Know) * 05:03 - Andreessen Takes Bold Stand Against Moral Discernment * 07:09 - Tech World Decides Performative Cruelty May Have Gone Too Far * 10:47 - The Avatar of Societal Decay * 11:39 - Marc’s Technical Takes And Arguments Also Are Not G...

Podcast episode for Kimi K2 Thinking * 00:00 - Introduction * 00:29 - Introducing K2 Thinking * 02:03 - Writing Quality * 02:52 - Agentic Tool Use * 04:03 - Overall * 05:10 - Are Benchmarks Being Targeted? * 06:19 - Just As Good Syndrome * 06:58 - Reactions * 09:29 - Otherwise It Has Been Strangely Quiet The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts a...

Podcast episode for Variously Effective Altruism. * 00:00 - Introduction * 00:13 - Preventing Value Drift * 01:14 - Maximizing Good Makes People Look Bad * 02:20 - No We Have No Tuition * 03:25 - Will MacAskill and the Dangers of PR Focus * 11:41 - Tag, You’re It * 14:21 - Impact Philanthropy The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive new posts and support t...

A “From the Archives” episode, the full Immoral Mazes sequence. * 00:00:00 - Introduction * 00:07:07 - Moloch Hasn’t Won, By Zvi Mowshowitz, December 28th, 2019 * 00:07:23 - Meditations on Moloch * 00:12:50 - That’s Not How This Works, That’s Not How Any of This Works * 00:15:15 - Moloch’s Army: An As-Yet Unjustified But Important Note * 00:16:53 - Meditations on Elua * 00:21:41 - Perfect Competit...

Podcast episode for On Sam Altman’s Second Conversation with Tyler Cowen. * 00:00:00 - Introduction * 00:00:58 - On The Sam Altman Production Function * 00:02:05 - On Hiring Hardware People * 00:04:45 - On What GPT-6 Will Enable * 00:10:30 - On government backstops for AI companies * 00:15:51 - On monetizing AI services * 00:21:16 - On AI’s future understanding of intangibles * 00:26:55 - On Chip-...

Podcast episode for AI #141: Give Us The Money. * 00:00:00 - Introduction * 00:01:52 - Table of Contents * 00:04:46 - Language Models Offer Mundane Utility * 00:05:11 - Language Models Don’t Offer Mundane Utility * 00:06:47 - Huh, Upgrades * 00:07:46 - On Your Marks * 00:18:01 - Deepfaketown and Botpocalypse Soon * 00:20:44 - Fun With Media Generation * 00:21:01 - They Took Our Jobs * 00:21:28 - A...

Podcast episode for Anthropic Commits To Model Weight Preservation. * 00:00 - Introduction * 01:36 - What Anthropic Is Doing * 09:30 - Releasing The Weights Is Not A Viable Option * 11:04 - Providing Reliable Inference Can Be Surprisingly Expensive * 14:03 - The Interviews Are Influenced Heavily By Context * 19:56 - Others Don’t Understand And Think This Is All Deeply Silly The Don’t Worry About t...

Podcast episode for OpenAI: The Battle of the Board: Ilya’s Testimony * 00:00 - Introduction * 01:06 - This Really Was Primarily A Lying And Management Problem * 03:25 - Ilya Tells Us How It Went Down And Why He Tried To Do It * 05:36 - If You Come At The King * 06:52 - Enter The Scapegoats * 07:37 - And In Summary The Don’t Worry About the Vase Podcast is a listener-supported podcast. To receive...