INDEX

Explanations

phrases related to news headlines and reports with an emphasis on the phrase "JUST WATCHED"

oai_token-act-pair · gpt-3.5-turbo

phrases indicating that something has been watched or viewed

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 situ

-0.73

 denomin

-0.68

duct

-0.67

 dissolution

-0.67

estate

-0.65

separ

-0.65

 dissolved

-0.65

cuts

-0.64

urus

-0.63

reach

-0.61

POSITIVE LOGITS

 WATCHED

1.79

 EntityItem

0.96

ichick

0.91

enegger

0.89

 Transcript

0.88

 ItemLevel

0.86

ï

0.84

 WATCH

0.81

dain

0.81

 Cosponsors

0.80

Activations Density 0.009%

phrases related to news headlines and reports with an emphasis on the phrase "JUST WATCHED"

phrases indicating that something has been watched or viewed

No Comments

No Known Activations