INDEX

Explanations

Q: followed by phrase

The neuron fires on technical identifiers—especially software/library names, acronyms, DSL keywords, and other “proper noun–style” tokens in code or documentation.

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

 dichter

-1.17

 léka

-1.17

älter

-1.16

laundry

-1.16

in

-1.15

ittäin

-1.14

работ

-1.12

boil

-1.09

炆

-1.08

עים

-1.07

POSITIVE LOGITS

of

1.60

 about

1.43

vo

1.33

1.28

ЧИТА

1.27

1.26

くちゃ

1.24

1.19

cs

1.18

Kegiatan

1.16

Activations Density 0.026%