INDEX

Explanations

instances where individuals admit to something or make a confession

oai_token-act-pair · gpt-3.5-turbo

phrases expressing acknowledgment or confession of wrongdoing

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ighth

-0.76

asus

-0.74

 ILCS

-0.71

osi

-0.70

 Airl

-0.67

assic

-0.67

DragonMagazine

-0.67

prev

-0.67

rouse

-0.66

Connector

-0.64

POSITIVE LOGITS

 defeat

1.22

 guilt

1.17

 wrongdoing

1.11

 mistakes

1.03

 ignorance

0.98

 fault

0.93

 failings

0.88

 admitting

0.87

 shortcomings

0.86

 responsibility

0.85

Activations Density 0.088%

instances where individuals admit to something or make a confession

phrases expressing acknowledgment or confession of wrongdoing

No Comments

No Known Activations