INDEX

Explanations

qualifier followed by noun

This neuron fires strongly on abstract nominalizations—technical nouns ending in “-ion,” “-ment,” “-ance,” “-ity,” etc., like “performance,” “behavior,” or “properties.”

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

 have

-1.84

 which

-1.82

and

-1.57

 with

-1.55

-1.53

 encompassing

-1.52

 explaining

-1.51

 demonstrated

-1.50

 shown

-1.48

 imagined

-1.46

POSITIVE LOGITS

our

2.03

 kartun

1.62

 visse

1.61

 Füßen

1.59

TWO

1.59

 Secrétaire

1.58

the

1.55

ometries

1.53

 dimanche

1.48

 individuelles

1.46

Activations Density 0.271%