Brave New World -- hosted by Vasant Dhar

Ep 58: Sam Bowman on ChatGPT & Controlling AI

Mar 23, 2023

01:12:27

Creator website

Highlights

AI Chapters

Episode notes

I Think It's Starting to Get Used to Cheat on Essays

01:20

How to incentivize a machine to give you medical advice

01:18

Predicting the Next Word

01:20

What Skill Oversight Can Look Like in the Real World

04:26

Introduction

2min

Introducing Natural Language Processing (GPT) Expert Sam Bowman

3min

Can Artificial Neural Networks Do Anything at All?

2min

What Was the Turning Point in NLP?

4min

Using GPUs to Train Neural Networks

2min

Is There a Sense of Cleverness in the Formulation?

5min

What Is a Large Language Model and How Does It Work?

5min

JADGPT - You Can Use This Auto-Complete Task to Predict

1min

Is That Mouse a Cookie Story?

5min

10.

Are the Foundations Strong Enough to Build More General Purpose Intelligence?

5min

11.

The Limitations of a Transformer

3min

12.

What Are the Potential Harms of AI?

4min

13.

Is OpenAI Really Getting Better?

3min

14.

Alignment Problems

4min

15.

Scalable Oversight of AI Systems

3min

16.

What Skill Oversight Can Look Like in the Real World

4min

17.

Is There a Use Case for Language Models?

2min

18.

Using Models to Oversee Models

4min

19.

The Results of the JatchPT Modeling Experiment

2min

20.

Are There Some Guardrails That We Should Be Thinking About?

5min

21.

What's Next for GPD3?

3min

Artificial intelligence is awesome -- but should we fear it? How can we stay in charge? Sam Bowman joins Vasant Dhar in episode 58 of Brave New World to discuss the Control Problem -- and more.

Also check out: 1. Sam Bowman at NYU Courant, LinkedIn,Twitter and Google Scholar. 2. Measuring Progress on Scalable Oversight for Large Language Models -- Samuel R Bowman et al. 3. Herbert Simon, Harry Pople and Norbert Wiener. 4. Language Models as Agent Models -- Jacob Andreas. 5. The Alignment Problem: Machine Learning and Human Values — Brian Christian. 6. Human Compatible — Stuart Russell. 7. Thinking, Fast and Slow — Daniel Kahneman. 8. ImageNet Classification with Deep Convolutional Neural Networks --Alex Krizhevsky et al. Check out Vasant Dhar’s newsletter on Substack. Subscription is free!

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.