12 - AI Existential Risk with Paul Christiano

1

Introduction

00:00 • 2min

2

Si, Human Potential in Expectation

01:44 • 2min

3

How to Delegate Cognitive Work in a System?

03:24 • 3min

4

Ai Systems

05:57 • 3min

5

Are There Any Risks in the Future?

08:39 • 2min

6

Are There Risks That Are Ai Risks?

10:33 • 3min

7

Is Ai Progress Going to Be Really Rapid?

13:44 • 4min

8

The Time Scale Is Measured More Like Years Than a Decade

17:19 • 2min

9

Is That Akin to a Dogenous Growth Story?

19:08 • 2min

10

What's the Like, Key Difference Between Humans and Ai Systems?

20:53 • 4min

11

When We'll Get Very Capable AI to Change Gears a Little Bit?

24:38 • 4min

12

Is There a Complimentary Empirical Question?

28:27 • 4min

13

Is There a Road Map for a Line D K?

32:22 • 2min

14

Ai Progress

34:52 • 4min

15

Is There an Economy of Scale in Training Machine Learning Systems?

38:45 • 3min

16

Is All the Eyes on Ai Technology Changing the Story?

41:32 • 4min

17

Is There a Secret Source to Intelligence?

45:43 • 3min

18

Is the Rate of Progress in Softras Getting Better?

49:00 • 2min

19

Are Your Doing Well or Poorly?

51:09 • 3min

20

Is There a Neural Network?

53:49 • 3min

21

Scaling Up

56:20 • 6min

22

How to Measure the Rate of Progress in Computer Science?

01:02:10 • 2min

23

Do You Think Cove 19 Has Changed Your Beliefs?

01:03:49 • 4min

24

What's the Difference Between Super Intelligence and Superintelligence?

01:08:00 • 2min

25

Is There a Definitive Strategic Advantage?

01:10:10 • 4min

26

What Technical Problems Could Cause Axential Risk?

01:14:02 • 5min

27

Do You Really Want to Copy Paste a Strawberry?

01:19:21 • 3min

28

Carve Up Research Problems Differently From Sources of Doom

01:22:18 • 2min

29

Intent Alignment

01:24:15 • 4min

30

Optimizing for Doom?

01:27:57 • 3min

31

Is There a Human Decomposition?

01:30:47 • 2min

32

Is There a Decomposition of Intent Alignment?

01:33:15 • 2min

33

Is It a Ranking Over Policies?

01:34:58 • 3min

34

Is There a Difference Between Outer and Inner Alignment?

01:37:33 • 3min

35

The Right Way to Specialize on a Sub Problem

01:40:28 • 2min

36

Is There a Difference in Methodological Approaches to Machine Learning?

01:42:55 • 3min

37

Are You Like, Ooh, Here's What I Want?

01:45:31 • 2min

38

Is It the Right Thing to Do in Computer Science?

01:47:12 • 4min

39

How Much Do You Rely on Hindsight to Evaluate Behaviour?

01:51:31 • 4min

40

Is There a Way to Improve Imitative Reenforcement Learning?

01:55:04 • 4min

41

Are You Interested in Interpretability?

01:59:25 • 2min

42

How Does It Help in Outer Linear Type Settings?

02:01:06 • 5min

43

The Problem Is That You Can't Predict Humans

02:06:08 • 2min

44

How to Do a Deepeneral Net Work?

02:08:16 • 3min

45

How to Train a System Like This?

02:11:20 • 3min

46

Using a Whiteboard to Write Code

02:14:40 • 2min

47

The Factored Cognition Approach to a Learning Alignment

02:16:12 • 5min

48

How to Optimize a Reference Manual

02:21:28 • 3min

49

Ai Systems

02:24:45 • 2min

50

The Basic Ingredient That Doesn't Sound Sufficient

02:26:36 • 3min

51

Is There a Need for Decoupling?

02:29:44 • 2min

52

The Basic Organizing Framework for Doom

02:31:16 • 3min

53

How to Write a Simple Pseudo Programmable Programmable Object

02:34:00 • 2min

54

What Are the Biggest Disagreements?

02:36:07 • 4min

55

Is There a Coupling Between Deep Learning and Machine Learning?

02:40:14 • 3min

56

What Are Your Most Important Uncertainties?

02:43:04 • 3min

57

Is There Anything That I Shouldn't Have

02:45:39 • 4min