INDEX

Explanations

phrases indicating an acknowledgment of errors or mistakes

oai_token-act-pair · gpt-3.5-turbo

phrases related to guilt and truthfulness

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 accompany

-0.75

lde

-0.72

 accompanying

-0.72

 Flavoring

-0.72

elve

-0.70

 greets

-0.70

Cooldown

-0.69

ingle

-0.69

»Ĵ

-0.67

phabet

-0.66

POSITIVE LOGITS

 wrong

1.90

 Wrong

1.68

wrong

1.64

 incorrect

1.55

 mistake

1.42

 faulty

1.39

 wrongly

1.37

 mistaken

1.33

Wr

1.32

 misleading

1.32

Activations Density 0.667%

phrases indicating an acknowledgment of errors or mistakes

phrases related to guilt and truthfulness

No Comments

No Known Activations