INDEX

Explanations

actions like fix, integrate, modify, prove

The neuron fires on past‐tense or past‐participle action verbs—particularly those ending in “-ed”—that denote completed technical or procedural changes (e.g. patched, fixed, characterized, distributed, modified, edited, etc.).

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

 jusqu

-1.38

''

-1.26

to

-1.20

 archite

-1.20

 sempre

-1.19

GEO

-1.18

("

-1.16

 robes

-1.15

 incluso

-1.13

SYSC

-1.09

POSITIVE LOGITS

 with

1.98

 this

1.76

 that

1.62

 fewer

1.56

 such

1.44

 their

1.41

 these

1.37

 having

1.36

 more

1.35

 removing

1.34

Activations Density 0.198%