INDEX

Explanations

as per, may benefit

This neuron lights up on relatively rare or content-specific words—essentially picking out informative keywords rather than common 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

 this

-1.83

 would

-1.66

 these

-1.63

 when

-1.57

 which

-1.52

 those

-1.48

 really

-1.46

 where

-1.41

but

-1.38

 actually

-1.35

POSITIVE LOGITS

 TikTok

1.28

 uniquement

1.28

 سانتی

1.23

Lé

1.23

acie

1.23

‌

1.21

tvguidetime

1.21

 constate

1.20

 seulement

1.19

Activations Density 0.009%