INDEX

Explanations

words indicating ongoing actions

The neuron fires strongly on tokens of “-ing” verbs (gerunds or present participles), even when they’re split across subword tokens, often together with their following particles (e.g. “paying off,” “lining up”).

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

<bos>

-1.60

 these

-1.46

currently

-1.44

-1.38

 сьогодні

-1.38

 diesen

-1.36

 sonriente

-1.31

 been

-1.31

 чрезвы

-1.31

both

-1.30

POSITIVE LOGITS

 after

1.89

 instead

1.51

 when

1.45

 with

1.42

in

1.40

着手

1.30

,”

1.27

as

1.27

 during

1.25

фильм

1.23

Activations Density 0.090%