INDEX

Explanations

references to negative events or situations

oai_token-act-pair · gpt-3.5-turbo

the word "Bad" used in various contexts related to failure or negative evaluations

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

cale

-0.83

 pulse

-0.66

rouse

-0.65

imate

-0.63

voy

-0.63

eph

-0.62

alysed

-0.62

forth

-0.61

cyl

-0.61

 incorpor

-0.61

POSITIVE LOGITS

Bad

3.79

Bad

2.92

BAD

2.23

bad

2.14

bad

1.69

Sad

1.44

 Good

1.33

 Poor

1.16

 Badge

1.15

 Evil

1.15

Activations Density 0.014%

references to negative events or situations

the word "Bad" used in various contexts related to failure or negative evaluations

No Comments

No Known Activations