INDEX

Explanations

contractions and abbreviations

The neuron is primarily detecting numeric tokens and sequences (i.e. numbers, years, and other digit‐based terms).

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 which

-1.92

if

-1.77

 then

-1.73

 while

-1.68

for

-1.67

MAXN

-1.66

to

-1.60

one

-1.57

 whose

-1.53

 your

-1.50

POSITIVE LOGITS

cluding

1.52

zość

1.52

 flesta

1.48

 compléter

1.48

腘

1.47

MANY

1.45

 MULTIPLE

1.43

BEFORE

1.41

Мы

1.40

黢

1.40

Activations Density 0.122%