INDEX

Explanations

phrases related to criminal activities or wrongdoings

oai_token-act-pair · gpt-3.5-turbo

references to crimes, particularly concerning violations against humanity

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

IFrame

Link

Negative Logits

alg

-0.80

rition

-0.71

ulously

-0.68

aran

-0.67

zl

-0.61

Dak

-0.61

 Gork

-0.60

ihara

-0.60

ulous

-0.60

bid

-0.59

POSITIVE LOGITS

 crimes

1.00

 Crimes

0.96

 offences

0.90

 offenses

0.87

 perpetrated

0.85

 unfocusedRange

0.84

 offenders

0.83

 Victim

0.81

 spree

0.80

 offend

0.79

Activations Density 0.019%

phrases related to criminal activities or wrongdoings

references to crimes, particularly concerning violations against humanity

No Known Activations