INDEX

Explanations

associations and their subsequent properties

The neuron fires on single words that signal a specific category or relation—especially those used in compound or hyphenated terms (e.g. “Nazi-related,” “gene variant,” “copyright owner,” “school districts”).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

ется

0.59

他也

0.57

 douze

0.57

 heller

0.55

 duodenum

0.53

시켜

0.52

 lệ

0.52

 Tomato

0.52

 zwölf

0.52

ും

0.51

POSITIVE LOGITS

以外

0.65

 olmayan

0.61

related

0.61

でない

0.61

aniyam

0.60

 berkaitan

0.60

 সম্পর্কিত

0.58

istin

0.58

 النهايات

0.58

 bothers

0.58

Activations Density 0.025%