Ep 17 : Andrej Karpathy (Tesla - AI) and Lex Fridman

1

Introduction

00:00 • 2min

2

Clip and Dally - What's the Biggest Breakthrough?

01:50 • 3min

3

Is There a Way to Train Convolutional Networks?

04:59 • 6min

4

Machine Learning Tooling Is Getting More Convenient

10:53 • 2min

5

Is It a Good Idea to Use Natural Language in Machine Learning?

12:24 • 4min

6

Language Modeling - A Massive Multitask Problem

16:22 • 2min

7

I Think It's a Good Idea to Do a Little Bit of Research on the Hardware Side of Things.

18:25 • 5min

8

The Role of Academic Labs in Open AI Research

23:54 • 4min

9

Is Data King of Machine Learning?

27:29 • 2min

10

Creating New Data Sets Is the Most Fun Part

29:15 • 3min

11

Transformers and Dally - What's the Difference?

32:21 • 3min

12

Transformers - The Greatest Substrat in NLP

35:34 • 4min

13

Is the Future of GPT-3 a Transformer?

39:16 • 5min

14

A Data Set Problem and a Social Problem in Machine Learning Research?

43:46 • 3min

15

Is There a Competition for Machine Learning?

47:06 • 1min

16

Is Multitask Learning the Next Big Barrier to Lifelong Learning?

48:32 • 5min

17

Is This a Good Way Forward?

53:29 • 5min

18

Is There an Intersection of Systems and Machine Learning?

58:02 • 4min

19

The Quest for Explainability in AI

01:01:37 • 1min

20

I Think Multimodal Explanations Could Be a Really Interesting New Research Area

01:02:48 • 5min

21

Is Differential Privacy a Good Idea?

01:07:52 • 3min

22

Deep Learning Systems in the Wild?

01:11:19 • 3min

23

Is There a Right IDE for Data Curation?

01:14:30 • 3min

24

Is There a Way to Accelerate Machine Learning?

01:17:44 • 2min

25

Are You Using Machine Learning to Predict Accuracy?

01:20:05 • 6min

26

Creating an AGI System?

01:26:05 • 2min

27

How to Combine AGI Logic and Probabilistic Statistical Reasoning?

01:28:13 • 3min