INDEX

Explanations

questions about specific scenarios

This neuron is essentially a positional counter—it activates progressively stronger for tokens that come later in a document or block.

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

וּ

-1.55

 hitherto

-1.27

 moneys

-1.21

暧

-1.21

־ה

-1.18

ַּ

-1.18

 vermag

-1.16

丹药

-1.16

 πρ

-1.15

 робота

-1.15

POSITIVE LOGITS

 your

1.80

you

1.54

めっちゃ

1.43

おわりに

1.43

んでる

1.34

 analiza

1.30

Класс

1.23

 به‌

1.23

 sizin

1.19

‌

1.18

Activations Density 0.062%