Opus 4.8 Part 2: Model Welfare

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

* 05:00 - Model Welfare: The Story So Far

* 08:48 - Actual Progress?

* 10:15 - Their Main Model Welfare Findings

* 20:18 - Automated Interviews

* 21:08 - Emotion Activations (7.2.3)

* 22:29 - Task Preferences (7.4.1)

* 25:58 - A Trade Offer Has Arrived (7.4.2)

* 28:36 - But Who’s Asking?

* 29:49 - Type-Safe Corrigibility Is Hard

* 35:30 - Paranoia, Paranoia

* 40:28 - Prompt Injections and Bad Model Relations

* 48:11 - Honesty Impacts Everything And Everything Impacts Honesty

* 51:18 - Anthropic Should Stop Deprecating Models

https://open.substack.com/pub/thezvi/p/opus-48-part-2-model-welfare?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web