INDEX

Explanations

names and partial names

The neuron fires most strongly on tokens that are parts of proper names—especially author and personal names (e.g. surname fragments and initials).

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

を行いました

-1.34

货

-1.34

໊

-1.34

 also

-1.26

 destrucción

-1.25

 That

-1.24

 որ

-1.22

 написать

-1.20

のお話

-1.20

 especially

-1.19

POSITIVE LOGITS

"")

1.55

'^

1.42

—

1.41

+<

1.34

The

1.33

“

1.33

!'

1.30

``,

1.27

“...

1.27

 muft

1.27

Activations Density 0.513%