INDEX

Explanations

words ending in "-men" or "-man."

oai_token-act-pair · gpt-3.5-turbo

references to males or male groups in various contexts

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 balance

-0.70

 outgoing

-0.65

 exception

-0.62

 energetic

-0.59

 unexpected

-0.59

 cheek

-0.59

 Wiley

-0.56

 identifiable

-0.55

 prism

-0.54

 overw

-0.54

POSITIVE LOGITS

men

4.42

mens

2.44

MEN

2.41

man

2.40

women

2.37

woman

2.24

manship

2.09

mans

2.08

Men

1.82

people

1.81

Activations Density 0.011%

words ending in "-men" or "-man."

references to males or male groups in various contexts

No Comments

No Known Activations