INDEX

Explanations

file path or code snippet

The neuron consistently lights up on standalone tokens that name files, extensions, or code identifiers (e.g. “MAX,” “html,” “script,” “input,” “avi,” “string,” “Images,” “Edition”), so it’s detecting file‐ or code‐related tokens.

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

for

-1.58

So

-1.49

 This

-1.47

-1.41

-1.36

For

-1.34

the

-1.29

so

-1.27

 There

-1.27

POSITIVE LOGITS

Vra

1.56

Tahun

1.43

Cevap

1.37

Recept

1.36

Lokasi

1.35

Zdrav

1.35

Makanan

1.35

Frage

1.32

Vegeta

1.32

 poêle

1.28

Activations Density 0.035%