INDEX

Explanations

the word "Are" at the beginning of questions

oai_token-act-pair · gpt-3.5-turbo

questions starting with "Are" that probe for confirmation or clarification

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.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

 hand

-0.57

 drafts

-0.57

 given

-0.55

 fixture

-0.55

 straight

-0.54

 dash

-0.54

 solution

-0.53

rious

-0.53

åĤ

-0.53

 priv

-0.53

POSITIVE LOGITS

Are

3.21

Are

2.12

 Were

2.02

 Aren

1.95

Is

1.84

ARE

1.68

Did

1.59

 Have

1.49

Were

1.48

Can

1.46

Activations Density 0.011%

the word "Are" at the beginning of questions

questions starting with "Are" that probe for confirmation or clarification

No Comments

No Known Activations