INDEX

Explanations

keywords related to misconduct, especially in legal or professional contexts

oai_token-act-pair · gpt-3.5-turbo

references to police and general misconduct issues

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

 Scan

-0.69

echo

-0.68

arest

-0.62

Loc

-0.61

 echo

-0.61

 optim

-0.61

-0.60

Loc

-0.59

 Lets

-0.59

Tag

-0.59

POSITIVE LOGITS

 misconduct

3.52

 wrongdoing

1.93

 malf

1.69

 negligence

1.38

 Misc

1.37

 criminality

1.32

 harassment

1.30

 wrongful

1.29

 fraud

1.29

 disciplinary

1.27

Activations Density 0.015%

keywords related to misconduct, especially in legal or professional contexts

references to police and general misconduct issues

No Comments

No Known Activations