INDEX

Explanations

words related to illegal or unethical behavior related to trading or sharing confidential information

oai_token-act-pair · gpt-3.5-turbo

references to "insider" and related concepts

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

anty

-0.68

nda

-0.65

 Grants

-0.64

ood

-0.64

enance

-0.63

ead

-0.63

à¤

-0.63

ez

-0.62

 Parenthood

-0.62

 Mankind

-0.62

POSITIVE LOGITS

 insiders

1.07

 insider

1.04

holders

0.78

aucus

0.75

iquid

0.69

wcs

0.68

holder

0.68

 workings

0.68

 informants

0.67

obin

0.64

Activations Density 0.006%

words related to illegal or unethical behavior related to trading or sharing confidential information

references to "insider" and related concepts

No Comments

No Known Activations