INDEX

Explanations

shower designs

The neuron fires on content-bearing, relatively rare or domain‐specific tokens (e.g. key nouns, names, numbers, or technical terms) rather than on frequent function words.

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

 what

-1.13

 where

-0.97

})$}

-0.96

 аналоги

-0.95

 when

-0.93

by

-0.92

 similar

-0.91

our

-0.90

 okuyayım

-0.88

 whatever

-0.85

POSITIVE LOGITS

着脸

1.03

 réinitial

0.95

鐽

0.93

 Créer

0.91

 rêver

0.91

 finir

0.91

 חיצוני

0.89

 rafra

0.88

 Trouver

0.87

 astuces

0.87

Activations Density 0.001%