INDEX

Explanations

symbols and punctuation

The neuron strongly activates on programming‐style symbols and punctuation (e.g. colons, double-colons, ampersands, file‐extension dots, variable sigils) rather than natural language words.

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

↵

-1.48

Hva

-1.41

 Når

-1.26

 prohi

-1.24

みんな

-1.23

५

-1.22

 facili

-1.20

饃

-1.20

 Hvordan

-1.20

WHILE

-1.20

POSITIVE LOGITS

in

2.41

for

2.38

 also

2.16

to

1.83

on

1.81

at

1.68

of

1.67

 other

1.57

 Also

1.54

or

1.53

Activations Density 0.124%