INDEX

Explanations

year, comma, then data

The neuron fires strongly on the short, usually single‐character tokens that appear as command-line option flags (e.g. “-B”, “-C”, “-d”, etc.) in a program’s usage/help message.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 কি

0.61

ạ

0.56

 მარ

0.55

 Fär

0.52

 correspond

0.51

 Kasım

0.51

 فبراير

0.50

只有一个

0.50

ewe

0.50

 பெண்கள்

0.49

POSITIVE LOGITS

xxxxx

1.15

XXXX

1.06

xxx

1.02

xxxx

1.01

XXXXXXXX

0.98

XXX

0.97

XXX

0.96

xxx

0.93

NAME

0.93

某

0.93

Activations Density 0.057%