INDEX

Explanations

names after Q, or titles after [, or names followed by names, or action starts writing

The neuron fires on substantive content words in legal‐case narratives (e.g. parties, key actors, and their factual actions) rather than on boilerplate or citations.

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

our

-1.78

 even

-1.76

we

-1.62

 will

-1.61

of

-1.41

you

-1.39

\}$,

-1.38

if

-1.33

as

-1.30

 just

-1.30

POSITIVE LOGITS

ַּ

1.87

ֶּ

1.75

ִּ

1.55

ְּ

1.52

ָּ

1.47

ּוֹ

1.47

 السياس

1.37

nappropriate

1.28

became

1.27

ighed

1.26

Activations Density 0.032%