INDEX

Explanations

names and titles

The neuron selectively activates on the initial subword tokens of capitalized proper names (i.e. named‐entity beginnings).

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

for

-1.55

[]:

-1.50

of

-1.47

 began

-1.45

you

-1.40

in

-1.34

 następujące

-1.31

 since

-1.29

so

-1.27

 like

-1.23

POSITIVE LOGITS

The

1.63

 applau

1.38

 divertir

1.27

 fré

1.27

该如何

1.27

 diriger

1.25

 ralenti

1.20

 evangé

1.20

 conster

1.18

 patienter

1.17

Activations Density 0.090%