Paul Christiano — Preventing an AI takeover

What do we want post-AGI world to look like?

0:00

Timelines

24:25

Evolution vs gradient descent

45:28

Misalignment and takeover

54:53

Is alignment dual-use?

1:17:23

Responsible scaling policies

1:31:38

Paul’s alignment research

1:58:25

Will this revolutionize theoretical CS and math?

2:35:01

How Paul invented RLHF

2:46:11

Disagreements with Carl Shulman

2:55:10

Long TSMC but not NVIDIA

3:01:53