INDEX

Explanations

negative adjectives and insults

oai_token-act-pair · gpt-3.5-turbo

derogatory terms aimed at individuals and their characteristics

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

iscover

-0.71

 ideally

-0.69

 outdoors

-0.68

Enc

-0.67

 incorporating

-0.67

arger

-0.65

 incorporate

-0.65

 Fold

-0.63

 consulted

-0.63

 conserv

-0.63

POSITIVE LOGITS

 pathetic

2.20

 worthless

1.89

 idiots

1.89

 stupidity

1.89

 hypocritical

1.86

 meaningless

1.83

 bullshit

1.83

 laughable

1.82

 pointless

1.81

 shitty

1.79

Activations Density 0.090%

negative adjectives and insults

derogatory terms aimed at individuals and their characteristics

No Comments

No Known Activations