The Gradient: Perspectives on AI

Catherine Olsson and Nelson Elhage: Anthropic, Understanding Transformers

Aug 26, 2022

47:04

Creator website

Episode guests

Nelson Elhage

Catherine Olsson

Highlights

AI Chapters

Episode notes

What Is in Context Learning and Induction Heads?

01:51

Getting Traction in the MLP Layers

03:44

The Basics of Attention Only Models

04:46

Introduction

2min

Anthropics

3min

Anthropic and Interpretability - What's Next?

3min

Is Mechanistic Interpretability a Good Idea?

5min

You've Got a Program on Mechanistic Interpretability

5min

The OV Matrix and the Path Expansion Frame

2min

Toy Transformers - How Do You Even Try to Do This?

2min

The Basics of Attention Only Models

5min

What Are Induction Heads?

2min

10.

Induction Heads

2min

11.

What Is in Context Learning and Induction Heads?

2min

12.

The Intact Context Learning Score

3min

13.

Induction Heads in Larger Models - The Key Takeaway

4min

14.

Getting Traction in the MLP Layers

4min

15.

How to Replicate the Induction Heads Findings

2min

16.

The Gradient Podcast - Part 2

2min

In episode 40 of The Gradient Podcast, Andrey Kurenkov speaks to Catherine Olsson and Nelson Elhage.

Catherine and Nelson are both members of technical staff at Anthropic, which is an AI safety and research company that’s working to build reliable, interpretable, and steerable AI systems. Catherine and Nelson’s focus is on interpretability, and we will discuss several of their recent works in this interview.
Follow The Gradient on Twitter

Outline:

(00:00) Intro
(01:10) Catherine’s Path into AI
(03:25) Nelson’s Path into AI
(05:23) Overview of Anthropic
(08:21) Mechanistic Interpretability
(15:15) Transformer Circuits
(21:30) Toy Transformer
(27:25) Induction Heads
(31:00) In-Context Learning
(35:10) Evidence for Induction Heads Enabling In-Context Learning
(39:30) What’s Next
(43:10) Replicating Results
(46:00) Outro

Links:

Anthropic

Zoom In: An Introduction to Circuits

Mechanistic Interpretability, Variables, and the Importance of Interpretable Bases

A Mathematical Framework for Transformer Circuits

In-context Learning and Induction Heads

PySvelte

Get full access to The Gradient at thegradientpub.substack.com/subscribe

Remember Everything You Learn from Podcasts

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