INDEX

Explanations

gender-specific words and titles, particularly focusing on the male gender

oai_token-act-pair · gpt-3.5-turbo

references to gender, race, and political affiliations

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.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

•Male Descriptors

No Comments

Negative Logits

utra

-0.81

cellence

-0.73

ruction

-0.71

alach

-0.71

WAY

-0.69

WAYS

-0.68

akens

-0.62

FontSize

-0.61

inventoryQuantity

-0.60

adobe

-0.59

POSITIVE LOGITS

 counterparts

2.06

 brethren

1.94

 counterpart

1.88

 cousins

1.81

 colleagues

1.79

 neighbors

1.74

 peers

1.73

 neighbours

1.72

 predecessors

1.71

 buddies

1.70

Activations Density 0.381%

gender-specific words and titles, particularly focusing on the male gender

references to gender, race, and political affiliations

No Comments

No Known Activations