INDEX

Explanations

Overall rating, Google Alert, Click here

This neuron detects section-introducing or UI/metadata keywords (like headings, prompts, and labels) that start lines or signal navigation (e.g., “Click,” “Extract,” “Assistance,” “Telegrams,” “Overall,” “Ladies”).

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

儀式

-1.38

Halaman

-1.32

voer

-1.27

Lá

-1.24

 statunit

-1.20

 katk

-1.15

Скачать

-1.14

小物

-1.13

Ilustra

-1.13

 Lainnya

-1.13

POSITIVE LOGITS

際に

1.40

様な

1.34

 klachten

1.28

ようです

1.23

ようで

1.23

様に

1.21

 ebenfalls

1.18

清晨

1.16

addCriterion

1.15

denly

1.15

Activations Density 0.025%