INDEX

Explanations

The neuron consistently lights up on the first non‐function (“content”) word at the start of each text segment.

past tense verbs describing actions

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

為に

-1.25

ходите

-1.25

-1.22

に行く

-1.21

På

-1.15

んですが

-1.14

 estos

-1.13

 enchufe

-1.13

 görüntüleri

-1.10

 dingen

-1.09

POSITIVE LOGITS

to

1.22

 alip

1.17

<bos>

1.16

 tass

1.08

Procedimento

1.07

леку

1.07

日記

1.07

 vila

1.06

💢

1.05

 $(\%)$

1.04

Activations Density 0.003%