INDEX

Explanations

hyphenated phrases or quoted attributions

The neuron fires on isolated punctuation “divider” tokens—especially dashes (– or -) and straight or curly quotation marks—i.e. characters used to set off or separate parts of the text.

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

 любых

-1.62

 sogar

-1.62

Даже

-1.59

Oft

-1.57

 حتی

-1.56

 любого

-1.52

与此同时

-1.48

Однако

-1.48

Если

-1.47

 arrog

-1.46

POSITIVE LOGITS

was

1.95

 they

1.89

on

1.80

 their

1.78

 also

1.67

 will

1.66

 finally

1.64

 imprensa

1.64

for

1.54

now

1.52

Activations Density 0.001%