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://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
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.