INDEX

Explanations

This neuron selectively fires on words appearing in long, continuous, formal‐style narrative or descriptive passages rather than short, casual snippets.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

是从

1.30

是为了

1.12

是用

1.12

 megawatts

1.10

तून

1.07

وية

1.07

 insurgents

1.05

 phonons

1.03

 দীননাথ

1.03

urier

1.01

POSITIVE LOGITS

position

1.40

 debout

1.40

1.36

in

1.30

 atop

1.20

 रहकर

1.20

1.18

ك

1.17

taught

1.14

faithful

1.13

Activations Density 4.441%