INDEX

Explanations

if you

The neuron fires on domain‐style tokens and other standout topic labels or jargon—i.e. uncommon, content-specific keywords (like “.org” domains, gene markers, procedures, CLI commands, section headings, etc.).

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

 内衣

-1.55

 estadounid

-1.52

bindet

-1.52

 花纹

-1.48

 手绘

-1.48

 素描

-1.40

 dáms

-1.35

Լ

-1.34

 短发

-1.32

 incrí

-1.30

POSITIVE LOGITS

you

2.02

 ever

1.81

any

1.64

 were

1.55

is

1.54

 want

1.52

 fails

1.52

or

1.49

 hadn

1.48

not

1.42

Activations Density 0.090%