INDEX

Explanations

references to individuals being convicted of crimes

oai_token-act-pair · gpt-3.5-turbo

instances of the word "convicted" in various contexts related to legal offenses

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

arity

-0.78

zl

-0.68

ww

-0.67

RIS

-0.67

 holiest

-0.66

opian

-0.65

Dak

-0.62

place

-0.62

lag

-0.62

anyl

-0.60

POSITIVE LOGITS

 convicted

1.07

 felon

0.99

 convict

0.91

 sentenced

0.85

 indicted

0.80

 convictions

0.80

 guilty

0.74

 jailed

0.72

 wiret

0.72

avorite

0.71

Activations Density 0.013%

references to individuals being convicted of crimes

instances of the word "convicted" in various contexts related to legal offenses

No Comments

No Known Activations