Laura Weidinger: Ethical Risks, Harms, and Alignment of Large Language Models

1

Introduction

00:00 • 4min

2

Is There a Difference Between AI and Cognitive Science?

03:40 • 3min

3

How Did You Become Interested in AI Ethics?

06:23 • 2min

4

How Do We Measure How Ethical AI Is?

08:50 • 3min

5

Foresight, Sandboxing, Participation

12:03 • 2min

6

Is There a Cultural Shift in AI Research?

14:28 • 3min

7

The Ethical and Social Risks of Harm From Language Models

17:17 • 3min

8

DeepMind, What Are the Types of Risks?

19:58 • 4min

9

The Impact of Language Models in AI Development

24:25 • 3min

10

How Do We Measure Harmful Text?

27:33 • 5min

11

What's the Main Insight From This Higher Level Synthesis?

32:45 • 3min

12

Is the AI Act Going to Be a Live Process?

35:36 • 3min

13

AI Ethics - You Can't Explain Deep Learning?

38:27 • 3min

14

Mispecification in Language Models

41:31 • 2min

15

Is There Any Way to Mimulate Mispecification From Training Data?

43:23 • 3min

16

GPT 4chan - What Are the Main Harms of Training GPTJ on Hate Speech?

46:07 • 2min

17

Are You Being Responsible When You Release a Model?

48:29 • 2min

18

Is 4chan a Hacker's Model?

50:33 • 2min

19

What Do You Do Outside of Research?

52:30 • 3min