Inefficiencies in training

Introduction

0:00

Scaling

1:00

Language

15:46

Economic Usefulness

22:58

Bioterrorism

38:05

Cybersecurity

43:35

Alignment & mechanistic interpretability

47:19

Does alignment research require scale?

57:43

Misuse vs misalignment

1:05:30

What if AI goes well?

1:09:06

China

1:11:05

How to think about alignment

1:15:11

Is modern security good enough?

1:31:31

Anthropic’s Long Term Benefit Trust

1:45:53

Is Claude conscious?

1:51:18

Keeping a low profile

1:56:14