INDEX

Explanations

words related to secrecy, including terms like "secret," "warrants," and "court orders."

oai_token-act-pair · gpt-3.5-turbo

references to secretive practices or operations

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 5-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.5.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.5.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

annis

-0.79

older

-0.76

gaard

-0.74

thood

-0.74

oker

-0.74

odcast

-0.73

DR

-0.73

adjusted

-0.70

alone

-0.69

 Stick

-0.69

POSITIVE LOGITS

arial

1.18

ariat

1.09

 ballot

0.91

 secret

0.90

 unfocusedRange

0.82

 memos

0.81

 informant

0.80

 contractors

0.79

 documents

0.78

 contractor

0.77

Activations Density 0.021%

words related to secrecy, including terms like "secret," "warrants," and "court orders."

references to secretive practices or operations

No Comments

No Known Activations