INDEX

Explanations

phrases related to danger and warning

oai_token-act-pair · gpt-3.5-turbo

references to danger or harmfulness

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

via

-0.83

rix

-0.82

elle

-0.75

ILA

-0.75

ļéĨĴ

-0.73

ARCH

-0.72

angular

-0.72

roma

-0.72

ann

-0.72

arger

-0.71

POSITIVE LOGITS

 dangerous

1.11

 undermin

1.02

 endanger

1.00

 adolesc

0.91

danger

0.89

 danger

0.87

 hazardous

0.85

 mosqu

0.84

 dangers

0.83

 deadly

0.80

Activations Density 0.015%

phrases related to danger and warning

references to danger or harmfulness

No Comments

No Known Activations