INDEX

Explanations

proper nouns, particularly names

The neuron fires on subword pieces that form the ends of proper names and technical terms (e.g. the trailing “t” in “Berndt,” “z” in “Cuntz,” “ton” in “Gunton,” or the suffixes of scientific and chemical names like “ites,” “oids,” “ure,” “il”).

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

＊

-2.75

?”

-2.45

and

-2.39

 revolves

-2.36

）

-2.34

”:

-2.31

—

-2.27

at

-2.25

\_

-2.20

表紙

-2.19

POSITIVE LOGITS

蟥

2.75

毉

2.66

ൎ

2.55

 plupart

2.52



2.50

 entsprechende

2.48

.&

2.42

 wahre

2.41

 These

2.39

 When

2.38

Activations Density 0.118%