INDEX

Explanations

starting with specific words

The neuron specializes in spotting numeric tokens and related punctuation (e.g. digits, “+”, quoted line/column numbers) in structured or code‐like text.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 from

-1.52

 only

-1.40

for

-1.34

 usually

-1.16

 often

-1.16

now

-1.14

！

-1.13

so

-1.11

 завдання

-1.09

 being

-1.07

POSITIVE LOGITS

 見える

1.20

rably

1.08

鼉

1.07

 videa

1.06

engl

1.06

 verdaderos

1.06

 coû

1.05

を変更

1.04

Reparto

1.02

really

1.02

Activations Density 0.001%