INDEX

Explanations

words related to violent actions and resulting harm or danger

oai_token-act-pair · gpt-3.5-turbo

language that indicates harm, danger, or serious physical injury

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 Compass

-0.78

 Sprite

-0.70

 cylinders

-0.68

audi

-0.67

 Leaders

-0.66

 quotas

-0.66

\":

-0.66

soDeliveryDate

-0.64

clus

-0.63

ellen

-0.63

POSITIVE LOGITS

 harm

1.65

 injury

1.58

 bodily

1.38

 injuries

1.35

 anguish

1.33

 griev

1.31

 harms

1.30

 inconvenience

1.29

 damage

1.28

damage

1.26

Activations Density 0.337%

words related to violent actions and resulting harm or danger

language that indicates harm, danger, or serious physical injury

No Comments

No Known Activations