INDEX

Explanations

conjunctions and clauses

The neuron fires strongly on the very last content word before a text‐block or line break, effectively marking the end of a segment.

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

 wrze

-1.47

ebly

-1.31

opies

-1.25

 Herzegovina

-1.23

닌

-1.22

athered

-1.21

そのため

-1.16

drawSprites

-1.11

rfan

-1.11

少爷

-1.11

POSITIVE LOGITS

and

4.56

 그리고

1.58

but

1.41

Karakter

1.27

AND

1.27

 descansar

1.23

Politik

1.14

direkt

1.13

There

1.11

 এবং

1.10

Activations Density 0.043%