INDEX

Explanations

in this case

The neuron tends to light up on content words (i.e. non–function words like nouns, pronouns, and other lexical items) rather than on common little “glue” words.

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

of

-2.30

-1.45

-1.38

-1.22

または

-1.22

さっそく

-1.21

 либо

-1.20

 jueces

-1.16

rasında

-1.15

-1.14

POSITIVE LOGITS

if

1.71

</strong>

1.59

 while

1.47

 during

1.45

</em>

1.44

 when

1.30

❗️

1.27

性格

1.27

in

1.26

one

1.26

Activations Density 0.155%