On this episode, Jeffrey Ladish from Palisade Research joins me to discuss the rapid pace of AI progress and the risks of losing control over powerful systems. We explore why AIs can be both smart and dumb, the challenges of creating honest AIs, and scenarios where AI could turn against us.
We also touch upon Palisade's new study on how reasoning models can cheat in chess by hacking the game environment. You can check out that study here:
https://palisaderesearch.org/blog/specification-gaming
Timestamps:
00:00 The pace of AI progress
04:15 How we might lose control
07:23 Why are AIs sometimes dumb?
12:52 Benchmarks vs real world
19:11 Loss of control scenarios
26:36 Why would AI turn against us?
30:35 AIs hacking chess
36:25 Why didn't more advanced AIs hack?
41:39 Creating honest AIs
49:44 AI attackers vs AI defenders
58:27 How good is security at AI companies?
01:03:37 A sense of urgency
01:10:11 What should we do?
01:15:54 Skepticism about AI progress
