INDEX

Explanations

suspicious individuals or people perceived as potential threats

oai_token-act-pair · gpt-3.5-turbo

references to suspects in criminal contexts

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

health

-0.67

lip

-0.66

 Eternity

-0.64

å¾

-0.61

arity

-0.60

utation

-0.59

birth

-0.59

 Vine

-0.58

 rejuven

-0.57

 roar

-0.57

POSITIVE LOGITS

 suspects

3.98

 suspect

2.34

 suspected

1.68

 perpetrators

1.56

 defendants

1.50

 offenders

1.47

 culprit

1.47

 detainees

1.39

 assailants

1.37

 arrests

1.32

Activations Density 0.010%

suspicious individuals or people perceived as potential threats

references to suspects in criminal contexts

No Comments

No Known Activations