INDEX

Explanations

date and time manipulation

The neuron is picking out programming‐style code blocks, especially curly‐brace tokens and indentation.

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

صلاح

-0.72

applicable

-0.71

tabela

-0.70

不思議な

-0.70

 pandemi

-0.69

örer

-0.68

ಲ

-0.66

endente

-0.66

halb

-0.65

imageURL

-0.65

POSITIVE LOGITS

Nj

0.81

쳤

0.81

 žena

0.77

yii

0.77

üsse

0.76

HDF

0.75

diğinde

0.75

premises

0.75

 Gedicht

0.75

doo

0.74

Activations Density 0.011%