INDEX

Explanations

keyword followed by common word

The neuron fires most strongly on key content words—especially domain-specific technical nouns or topic-central terms—marking important semantic keywords.

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

 godk

-1.66

Fritz

-1.58

 gangen

-1.52

 oeil

-1.51

 sociedades

-1.49

 もの

-1.45

Hilsen

-1.43

 özellikleri

-1.43

Läh

-1.42

 besk

-1.41

POSITIVE LOGITS

 same

1.70

 following

1.64

Its

1.63

1.41

 twenty

1.39

 number

1.37

of

1.37

 last

1.34

 forty

1.33

 question

1.31

Activations Density 0.198%