INDEX

Explanations

words related to a feeling of moral or factual incorrectness

oai_token-act-pair · gpt-3.5-turbo

statements indicating something is considered incorrect or morally wrong

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

rien

-0.66

Â¯Â¯Â¯Â¯

-0.65

ribes

-0.63

cit

-0.63

anned

-0.62

apple

-0.62

glas

-0.62

usters

-0.61

ility

-0.61

é¾

-0.60

POSITIVE LOGITS

 wrong

1.02

 unfocusedRange

0.89

wrong

0.86

 culprit

0.80

fully

0.77

ibrary

0.75

 mistaken

0.74

 Wrong

0.74

 tack

0.73

eous

0.70

Activations Density 0.014%

words related to a feeling of moral or factual incorrectness

statements indicating something is considered incorrect or morally wrong

No Comments

No Known Activations