INDEX

Explanations

areas, dictated, investigations, important questions, ways

The neuron fires on document sections that look like formal technical‐paper headings or abstract/introduction text (i.e. numbered or titled sections with discipline-specific terminology).

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

voer

-1.65

 jiné

-1.59

 sucht

-1.47

 variaciones

-1.43

 precies

-1.43

ämme

-1.42

 calabaza

-1.41

 semillas

-1.41

 effectué

-1.38

车子

-1.37

POSITIVE LOGITS

 required

1.77

亟

1.48

靭

1.42

靱

1.38

 design

1.34

 TAMBÉM

1.30

for

1.29

θούν

1.27

 иногда

1.27

 anello

1.22

Activations Density 0.205%