INDEX

Explanations

The neuron fires on code identifier name fragments—especially CamelCase or PascalCase tokens (e.g. interface/class/method/type name components like “ISomeOther,” “getTokenPrevious,” “setCongressMessage,” “signingMethod,” “Latest,” etc.).

class and type names

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

 τὸ

-1.01

 menant

-1.00

 THING

-0.95

 REGULATIONS

-0.94

 vantagens

-0.94

mby

-0.93

 dépens

-0.92

 manquante

-0.91

 příjem

-0.90

celand

-0.89

POSITIVE LOGITS

ton

0.95

 практически

0.85

王的

0.84

0.83

 testing

0.83

 явля

0.83

ca

0.82

 seek

0.82

 incluso

0.81

Activations Density 0.010%