INDEX

Explanations

via parenthetical examples

This neuron fires on structural/formatting cues and common Latin/academic connectors (e.g. section titles like “Abstract,” linking words such as “via,” and abbreviations like “i.e.” or “e.g.”).

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

 same

-1.41

 many

-1.25

all

-1.23

 this

-1.21

 best

-1.18

 sort

-1.15

 other

-1.15

 each

-1.14

 also

-1.12

 most

-1.09

POSITIVE LOGITS

</i>

1.70

</em>

1.60

 sice

1.14

 manquer

1.12

 zijne

1.10

Produkt

1.10

Vä

1.10

肮

1.09

strategie

1.09

Можно

1.09

Activations Density 0.002%