Episode 18: Oleh Rybkin, UPenn, on exploration and planning with world models

1

Introduction

00:00 • 5min

2

How Did You Get Into Your Pagete?

05:01 • 4min

3

Nour Science and the Architecture of the Brain

08:57 • 4min

4

I've Been Thinking About Reason

13:24 • 5min

5

Is There a Big Transformer in the Learning Process?

18:12 • 4min

6

How to Adapt to New Situations?

22:19 • 3min

7

Do You Have Intuition on How to Get Quick Adoption to New Situations?

24:55 • 1min

8

Morals Enable Latent Learning

26:13 • 6min

9

Plan to Explore

31:45 • 5min

10

The Best Way to Measure Uncertainty Estimation

36:18 • 2min

11

The Old Radio Isn't Working, Is It?

37:59 • 5min

12

Theorems of Theoeo Fasti

43:24 • 4min

13

How to Define Scales?

47:22 • 2min

14

Using a Robotic Agent

49:37 • 3min

15

The Agent With Four Legs

52:46 • 3min

16

The Right Way to Specify Goals

55:18 • 3min

17

How Do We Learn Language Conditional Agents?

58:25 • 2min

18

The Problem Is Learning From the Internet

01:00:37 • 2min

19

Is There an Alternative to Generalization in Reenforcement Learning?

01:02:59 • 4min

20

I Don't Understand Tiktok, I Don't Understand Tick Talk.

01:06:42 • 3min

21

Is There a Need for Better Generalization?

01:10:04 • 4min

22

How to Train a Moral and Image Nit for a Big Data Set

01:13:45 • 5min

23

How to Train a Dynamic Model on Utope?

01:19:06 • 3min

24

The Structure People Have Lost to the Giant Transformers in Conpee Mission

01:22:12 • 6min

25

Morals Are Important, Right?

01:27:54 • 4min

26

Is There a Benchmark for Adapting to a New Situation?

01:32:02 • 4min

27

Train Robert to Open a Cabinet in the Google Kitchen

01:35:52 • 3min

28

How to Train a Biped Robbert to Walk in Simulation

01:39:12 • 4min

29

The Future of Robotic Manipulation

01:43:19 • 5min

30

How to Come Up With a Good Idea

01:48:33 • 3min

31

The Way People Fail to Write Good Papers

01:51:08 • 3min

32

Did a Lot of Things Work?

01:54:26 • 4min

33

Plant Explore Does Work, but It's Still Too Simple of a Result

01:58:14 • 3min