INDEX

Explanations

if we

This neuron selectively lights up on high-frequency grammatical “function” words—common connectors such as if, has, it, once, course, anywhere, time, etc.—rather than on content words.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

在我

0.70

在他的

0.64

InBuffer

0.58

在他

0.58

在我们

0.58

जैसा

0.56

 कहाँ

0.55

的人生

0.55

ที่จะ

0.55

Surely

0.55

POSITIVE LOGITS

 they

1.93

 они

1.70

we

1.66

he

1.59

she

1.57

 вони

1.56

they

1.52

 তারা

1.41

you

1.32

 someone

1.31

Activations Density 0.294%