INDEX

Explanations

also searching

The neuron fires on common function words and discourse markers (e.g. articles, pronouns, adverbs), effectively spotting high‐frequency “stop” 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

and

-1.42

 nové

-1.30

 områ

-1.27

苺

-1.20

 zacht

-1.20

 hoger

-1.17

 hvem

-1.15

 olabilir

-1.15

-1.13

THEY

-1.12

POSITIVE LOGITS

 configureStore

1.23

 lampada

1.23

 potem

1.22

 tiende

1.20

 useAppContext

1.16

 coração

1.16

 coscienza

1.15

 setCode

1.15

 nede

1.14

())));

1.13

Activations Density 0.095%