INDEX

Explanations

core concept

The neuron strongly activates on abstract, nominalized concepts—especially multisyllabic nouns ending in “-tion” (e.g. “contradiction,” “development,” “juxtaposition”).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 bulletin

0.59

bullet

0.58

 훨씬

0.58

 bulletins

0.57

 Bulletin

0.56

 LinkedList

0.55

 ungroup

0.55

 unimportant

0.52

 grouped

0.51

 clustered

0.51

POSITIVE LOGITS

 contradiction

2.39

 contradictory

2.33

矛盾

2.23

 contradictions

2.20

 paradox

2.14

 conflicting

2.05

 paradoxical

2.05

contrad

2.03

 dichotomy

2.00

 contradict

1.88

Activations Density 0.571%