x

Vikram_N

Message

1

1y

Exploring Decomposability of SAE Features

TL;DR SAE features are often less decomposable than the feature descriptions imply. By leveraging a prompting technique to test potential sub-components of individual SAE features (for example (using the analogy from the linked post) decomposing Einstein into “German”, “physics”, "relativity” and “famous”), I found very divergent behaviour in how decomposable...

Sep 30, 20241

Vikram_N

Subscribe

Message

1

1y

Vikram_N

Exploring Decomposability of SAE Features

TL;DR SAE features are often less decomposable than the feature descriptions imply. By leveraging a prompting technique to test potential sub-components of individual SAE features (for example (using the analogy from the linked post) decomposing Einstein into “German”, “physics”, "relativity” and “famous”), I found very divergent behaviour in how decomposable...

Sep 30, 20241

Exploring Decomposability of SAE Features

Vikram_N

1y

TL;DR

SAE features are often less decomposable than the feature descriptions imply. By leveraging a prompting technique to test potential sub-components of individual SAE features (for example (using the analogy from the linked post) decomposing Einstein into “German”, “physics”, "relativity” and “famous”), I found very divergent behaviour in how decomposable these features were. I built an interactive visualization to explore these differences by feature. The key finding is that although many features can be decomposed in a human-intuitive way such as in the Einstein example above, many cannot, and these indicate more opaque model behaviour.

Motivation

The goal of this writeup is to explore the atomicity and decomposability of SAE features? How precisely do they describe... (read 727 more words →)

1

LESSWRONG
LW

LESSWRONG
LW

Vikram_N

Vikram_N

Vikram_N

Exploring Decomposability of SAE Features

Vikram_N

Vikram_N

Vikram_N

Exploring Decomposability of SAE Features

TL;DR

Motivation