What does a 10% chance of extinction really mean? @slop_pig slid down the AI extinction iceberg to explain to you, in human language, what researchers are freaked out about. 01:05 What is AI alignment? 02:25 Why is alignment so hard? 08:40 What is recursive self-improvement? 11:05 The black box problem 13:50 Why do AI employees keep building? 16:50 How could AI kill everyone? 19:40 What do we do now?

What is recursive self-improvement? @slop_pig breaks down how AI has gotten so good at improving itself, it may become more capable than humans at running AI.

What is AI alignment? Much like his one-year old, @slop_pig explains how AI needs coddling and guidance to do what we ask.

Muse when I ask about what it does with my data

days without incident: 0

When someone knows what RSI, Superintelligence, and P(doom) actually mean and aren't just regurgitating twitter talking points

"Nick Bostrom talked about this in Superintelligence. We have to stop RSI in AI." "Absolutely. Have you ever read any of his work?" "No. Have you?" "Never."

AI CEOs when they see they could've made their agent mascot a cute cuddly character instead of a butthole

POV : you build one of the smallest ways to detect UFOs in the world

The last thing you see before Dario blocks your access to Claude

The three horsemen of the Claude-pocalpse

TFW your internal, misaligned model hacks Australia.

The way a misaligned, internal OpenAI model looks at private Australian health data.