INDEX

Explanations

foreign language fragments

The neuron fires strongly on non-native or “foreign” tokens—especially Japanese katakana words and accented/irregular Latin-script fragments.

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

婠

-2.39

FundMe

-2.30

珰

-2.20

of

-2.19

嫒

-2.16

鸲

-2.11

⟞

-2.09

鹋

-2.06

ༀ

-2.06

ৄ

-2.00

POSITIVE LOGITS

 what

2.16

 although

1.95

Although

1.84

 However

1.84

After

1.81

What

1.79

</b>

1.77

but

1.77

Many

1.75

 both

1.74

Activations Density 0.005%