INDEX

Explanations

references to violence in media, particularly in cartoons and action-adventure genres

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

harmlessness

np_acts-logits-general · gemini-2.5-flash-lite

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GEMMA-2-9B-IT @ 20-gemmascope-res-131k

Configuration

google/gemma-scope-9b-it-res/layer_20/width_131k/average_l0_81

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Features

131,072

Data Type

float32

Hook Name

blocks.20.hook_resid_post

Hook Layer

Architecture

jumprelu

Context Size

1,024

Dataset

monology/pile-uncopyrighted

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

MemoryWarning

-0.54

AntiForgeryToken

-0.53

IntoConstraints

-0.53

WebServlet

-0.52

 Majefty

-0.50

HasBeenSet

-0.50

sizeCache

-0.47

IVEREF

-0.47

SequentialGroup

-0.47

 dAtA

-0.47

POSITIVE LOGITS

 harmless

0.75

 benign

0.62

 mild

0.60

 milder

0.58

 peaceful

0.56

 gentler

0.56

 gentle

0.55

mild

0.54

peaceful

0.53

Mild

0.50

Activations Density 0.152%

references to violence in media, particularly in cartoons and action-adventure genres

harmlessness

No Comments

No Known Activations