INDEX

Explanations

references to confidential information or secrets

oai_token-act-pair · gpt-3.5-turbo

the word "Secret" in various contexts related to secrecy or confidential matters

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

 grap

-0.76

Ö

-0.72

fits

-0.71

 displacement

-0.69

BCE

-0.69

 displ

-0.69

ettle

-0.67

bom

-0.66

orth

-0.66

 tackle

-0.63

POSITIVE LOGITS

 Secret

4.00

Secret

3.11

secret

2.27

 secret

2.09

 Secrets

2.00

 secrets

1.47

 Mysterious

1.32

 Hidden

1.23

 Postal

1.20

KGB

1.19

Activations Density 0.015%

references to confidential information or secrets

the word "Secret" in various contexts related to secrecy or confidential matters

No Comments

No Known Activations