INDEX

Explanations

conjunctions 'or' and 'and' signaling contrasting or additive relationships

oai_token-act-pair · gpt-3.5-turbo

phrases indicating moral comparisons between good and bad

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 11-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.11.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.11.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

ivari

-0.71

quit

-0.70

 LEVEL

-0.68

 veins

-0.68

igham

-0.66

brates

-0.66

ĸļ

-0.64

sbm

-0.64

igraph

-0.63

aturday

-0.63

POSITIVE LOGITS

 evil

1.13

bad

1.09

evil

1.08

 brightest

1.00

Evil

0.98

bad

0.89

 Evil

0.86

 wrong

0.84

 evils

0.83

BAD

0.83

Activations Density 0.099%

conjunctions 'or' and 'and' signaling contrasting or additive relationships

phrases indicating moral comparisons between good and bad

No Comments

No Known Activations