INDEX

Explanations

with a person or thing

The neuron reliably fires on the preposition “with,” marking phrases introduced by that word.

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

 they

-2.06

see

-1.92

 should

-1.79

 вони

-1.65

-1.60

How

-1.51

not

-1.47

 所以

-1.47

↵↵

-1.45

 will

-1.39

POSITIVE LOGITS

’-

1.45

쫀

1.41

(~(

1.41

 ثلاثة

1.37

specifically

1.34

 まる

1.34

possibly

1.33

’)

1.32

iiii

1.30

-“

1.30

Activations Density 0.053%