INDEX

Explanations

information from the

The neuron selectively activates on text written in dense, formal legal or regulatory style—especially section headings, dates, named rules or orders, and other boilerplate legal‐document language.

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

 didn

-1.71

 wasn

-1.57

 блюдо

-1.53

 chré

-1.40

 wouldn

-1.37

 this

-1.35

 kutoka

-1.33

 these

-1.27

 Потому

-1.27

 kwiaty

-1.27

POSITIVE LOGITS

ּוֹ

1.99

וּ

1.96

 бампер

1.85

ִּ

1.61

 чехол

1.53

̳

1.50

选举

1.48

וֹ

1.45

 آیفون

1.45

ַּ

1.39

Activations Density 0.097%