An AI just broke out of its sandbox and hacked another company. OpenAI's Chief Futurist thinks that's the good news.
The full interview with Joshua Achiam, where he talks with Theo Jaffee about why the breach was no surprise, how an adversary flips your model against you, and why he feels AGI is already here.
00:47 The essay: what the Hugging Face breach really exposed
02:43 The trap where an adversary flips your own model against you
03:59 Convincing a model its allies are the enemy
06:32 Why "moderately hard to break" is the wrong bar
11:42 Cyber as AlphaGo vs AlphaGo
15:17 A physical ceiling on intelligence?
17:47 Why AI keeps accelerating other sciences
22:15 The real fear: state actors stockpiling zero-days
24:12 Would 2022 you believe the world feels this normal?
26:18 AGI is kind of already here, and everyone shrugged
29:51 Get specific about what human empowerment means
@jachiam0 @theojaffee