INDEX

Explanations

names and specific entities

This neuron fires on isolated capitalized tokens, i.e. proper‐name parts.

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

Можно

-1.23

都已经

-1.22

多かった

-1.19

Technologie

-1.18

 DOMINGO

-1.18

Может

-1.17

 많은

-1.13

 بسیار

-1.13

症

-1.09

 diligent

-1.08

POSITIVE LOGITS

is

1.81

was

1.71

 didn

1.61

 would

1.55

 will

1.53

are

1.50

has

1.42

 wouldn

1.40

can

1.38

 couldn

1.37

Activations Density 0.106%