INDEX

Explanations

The neuron spikes on content‐heavy, relatively rare words (i.e. high‐information tokens) rather than common function words.

possessive pronouns and specific foreign 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

 your

-1.63

انظر

-1.59

You

-1.40

 their

-1.36

ätie

-1.35

־ה

-1.34

 goes

-1.33

We

-1.25

-1.24

 makes

-1.22

POSITIVE LOGITS

my

1.52

 kakao

1.43

 моей

1.43

kuy

1.42

 моих

1.28

kiz

1.26

Bí

1.22

 моего

1.22

siy

1.21

那样

1.20

Activations Density 0.005%