INDEX

Explanations

This neuron fires strongly on the very first meaningful token of each text segment, effectively picking out the “heading” or primary topic word at the start of a passage.

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

 entrust

1.32

 ascertain

1.27

 TRAILS

1.20

 advisers

1.20

 AGRE

1.17

 tremors

1.16

 appellees

1.14

 enchant

1.13

 amends

1.13

 auxili

1.13

POSITIVE LOGITS

larda

1.66

т

1.59

да

1.56

zelfde

1.55

ν

1.52

ği

1.50

estens

1.42

satz

1.41

rophic

1.39

kinson

1.38

Activations Density 13.798%