INDEX

Explanations

mentions of dangerous or harmful situations

oai_token-act-pair · gpt-3.5-turbo

instances of the word "deadly" in relation to various harmful events or conditions

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.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.3.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

erous

-0.87

agate

-0.87

ourced

-0.84

ership

-0.80

sama

-0.78

arity

-0.77

adra

-0.74

 Catalog

-0.74

estamp

-0.74

via

-0.74

POSITIVE LOGITS

 poisonous

0.97

 deadly

0.95

 wounding

0.87

 lethal

0.86

 poison

0.82

 assault

0.82

 violence

0.78

 threats

0.78

ãĥ£

0.78

 fatal

0.77

Activations Density 0.011%

mentions of dangerous or harmful situations

instances of the word "deadly" in relation to various harmful events or conditions

No Comments

No Known Activations