INDEX

Explanations

quoted strings or code delimiters

The neuron fires on isolated code‐style punctuation characters (commas, colons, parentheses, etc.) rather than on ordinary 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

 geliefert

-1.31

 الذي

-1.29

ńskich

-1.26

ńskiego

-1.24

 تُ

-1.23

 arcilla

-1.19

läg

-1.16

 impresionantes

-1.14

 يُ

-1.13

ensus

-1.13

POSITIVE LOGITS

 then

1.78

any

1.61

 cualquier

1.51

 such

1.44

 like

1.36

of

1.34

 qualsiasi

1.26

 cualquiera

1.25

 любой

1.21

ならば

1.16

Activations Density 0.114%