INDEX

Explanations

The neuron fires on tokens that are unusual proper‐noun or technical identifiers—especially multi-word company names, acronyms, all-caps words, or code/field names.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 ...]

0.64

itd

0.62

.}}

0.59

 achromatic

0.58

শনের

0.57

 cauza

0.57

 labyr

0.56

...?

0.55

အနေ

0.55

 chwili

0.55

POSITIVE LOGITS

",

2.55

")

2.42

").

2.23

"،

2.22

";

2.20

”,

2.16

");

2.14

"),

2.11

","

2.09

Activations Density 0.169%