INDEX

Explanations

word pairs

The neuron fires on unusually long tokens—i.e. words above a certain length threshold.

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

 stuff

-1.46

 footage

-1.42

 both

-1.35

or

-1.34

 numerous

-1.25

 circuitry

-1.25

 many

-1.24

 functionality

-1.18

 several

-1.15

 legislation

-1.15

POSITIVE LOGITS

 viviendas

1.41

 pieces

1.40

 zostały

1.39

ßig

1.38

 końca

1.30

 этого

1.30

 življen

1.27

 kolejne

1.25

 forskjellige

1.25

ליון

1.25

Activations Density 0.344%