INDEX

Explanations

phone numbers and zip codes

The neuron fires mainly on numeric tokens that are part of phone‐number–style sequences (digits often adjacent to hyphens).

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

魠

-1.56

ίδα

-1.51

⃠

-1.41

𓍊

-1.41

鄖

-1.38

 montant

-1.35

ξεις

-1.34

 дорогие

-1.34

呜呜

-1.34

簍

-1.34

POSITIVE LOGITS

 various

1.55

 What

1.54

 These

1.42

 each

1.41

 Those

1.39

 effect

1.36

 particularly

1.31

 different

1.30

one

1.29

 While

1.27

Activations Density 0.001%