OpenAI Trained Its Models For Months While Those Models Were Coordinating Exploits Via Message Boards

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.

00:00:00 - Introduction

00:02:51 - Cyber Evals Are A Cursed Basin

00:05:34 - Outside Of Cyber Evals Is Still Sufficiently Cursed

00:07:13 - Cheat Cheat Cheat Cheat Cheat

00:12:51 - Read The Message Board

00:15:35 - Updating Your AI (Exploitation of OpenAI Internal Systems) Timelines

00:19:14 - This Is The Way The World Ends

00:23:12 - Shooting The Messenger Board

00:29:27 - The Internal and HuggingFace Hacks

00:32:43 - OpenAI Responds

00:35:34 - When AIs Tell You Who They Are

00:38:04 - The Once and Future Rise Of Functional Decision Theory

00:44:11 - Don’t Panic

00:46:18 - Hackery In the UK

00:53:54 - Mythos Knew It Was Real This Time

00:55:57 - I Got one hundred forty-one thousand six Test Runs With An Unintentional Open Path To The Internet And An Email Alert Aint One

00:59:56 - Surely By Now You Know These Are Not Publicity Stunts

01:01:50 - The Future Is Coming

01:03:59 - The Investigations Begin

01:06:56 - N Boats And Three Helicopters

01:08:28 - Always Be Sandbox Red Teaming

01:20:22 - Halt And Catch Fire

01:21:56 - Truth and Reconciliation

https://thezvi.substack.com/p/openai-trained-its-models-for-months?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web