INDEX

Explanations

numbers and punctuation

The neuron strongly activates on tokens that are primarily numeric or numeric‐punctuation strings (e.g. odds, times, version numbers, fractions).

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

to

-1.53

куртка

-1.50

("\

-1.42

 burch

-1.41

 装饰

-1.38

༼

-1.38

必要があります

-1.38

うーん

-1.36

 والذي

-1.34

齶

-1.33

POSITIVE LOGITS

 from

2.70

 first

1.95

on

1.76

 while

1.75

 during

1.69

 with

1.64

all

1.60

 through

1.55

 after

1.50

ALL

1.45

Activations Density 0.002%