Training Models to Fool Monitors

Podcast episode for AI #147: Flash Forward.

* 00:00:00 - Introduction

* 00:01:38 - Table of Contents

* 00:05:13 - Language Models Offer Mundane Utility

* 00:07:09 - Language Models Don’t Offer Mundane Utility

* 00:07:59 - Huh, Upgrades

* 00:11:33 - On Your Marks

* 00:14:33 - Choose Your Fighter

* 00:17:57 - Get My Agent On The Line

* 00:22:56 - Deepfaketown and Botpocalypse Soon

* 00:25:55 - Fun With Media Generation

* 00:28:48 - Copyright Confrontation

* 00:29:15 - Overcoming Bias

* 00:31:16 - Unprompted Attention

* 00:32:10 - They Took Our Jobs

* 00:34:46 - Feeling the AGI Take Our Jobs

* 00:40:30 - The Art of the Jailbreak

* 00:40:50 - Get Involved

* 00:43:14 - Introducing

* 00:44:03 - Gemini Flash 3

* 00:46:29 - In Other AI News

* 00:47:06 - Going Too Meta

* 00:51:50 - Show Me the Money

* 00:52:28 - Bubble, Bubble, Toil and Trouble

* 00:54:39 - Quiet Speculations

* 00:56:46 - Timelines

* 00:57:46 - The Quest for Sane Regulations

* 01:00:51 - My Offer Is Nothing

* 01:09:16 - My Offer Is Nothing, Except Also Pay Me

* 01:13:41 - Chip City

* 01:20:44 - The Week in Audio

* 01:21:59 - Rhetorical Lack Of Innovation

* 01:33:12 - People Really Do Not Like AI

* 01:40:25 - Rhetorical Innovation

* 01:43:45 - Bad Guy With An AI

* 01:48:08 - Misaligned!

* 01:53:36 - Aligning a Smarter Than Human Intelligence is Difficult

* 01:56:27 - Mom, Owain Evans Is Turning The AIs Evil Again

* 02:03:17 - Messages From Janusworld

* 02:06:43 - The 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.

https://open.substack.com/pub/thezvi/p/ai-147-flash-forward?utm_campaign=post-expanded-share&utm_medium=web