INDEX

Explanations

news headlines indicating a video has been watched, typically paired with a phrase in all caps

oai_token-act-pair · gpt-3.5-turbo

instances of the word "watched" and its variations, indicating a focus on viewing or media consumption

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 10-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.10.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.10.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

etheless

-0.87

genre

-0.81

appropri

-0.79

soDeliveryDate

-0.75

fortunately

-0.75

ional

-0.72

ococ

-0.69

 comprom

-0.69

aten

-0.69

brainer

-0.69

POSITIVE LOGITS

 Highlights

1.23

 Inside

1.08

How

1.07

 Transcript

1.07

Why

1.06

 Timeline

1.06

 What

1.05

 WATCH

1.05

 Recap

1.02

 Coverage

1.01

Activations Density 0.027%

news headlines indicating a video has been watched, typically paired with a phrase in all caps

instances of the word "watched" and its variations, indicating a focus on viewing or media consumption

No Comments

No Known Activations