INDEX

Explanations

words related to secrecy, confidentiality, and hidden information

oai_token-act-pair · gpt-3.5-turbo

references to classified or confidential information

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

equ

-0.68

 ticks

-0.65

 averages

-0.65

 Jones

-0.64

 Burnett

-0.62

Cox

-0.62

 align

-0.61

 Fowler

-0.61

 alternatives

-0.60

lyn

-0.60

POSITIVE LOGITS

secret

4.21

Secret

2.36

 secret

1.76

 Secret

1.69

Secretary

1.33

 Secrets

1.23

hidden

1.20

 secretaries

1.20

 secrets

1.18

RET

1.17

Activations Density 0.018%

words related to secrecy, confidentiality, and hidden information

references to classified or confidential information

No Comments

No Known Activations