INDEX

Explanations

here's a breakdown

The neuron responds to formatting and boundary tokens—especially end‐of‐sentence punctuation, parenthetical markers, and blank‐line separators.

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

’’

0.96

 “”

0.94

?”

0.93

 ‘’

0.89

?’

0.88

’”

0.88

.’’

0.86

 ’’

0.84

,’’

0.84

 “‘

0.82

POSITIVE LOGITS

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

1.38

↵↵↵↵↵↵↵↵↵↵↵↵↵

1.35

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

1.34

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

1.30

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

1.27

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

1.25

↵↵↵↵↵↵↵↵↵↵↵

1.24

↵↵↵↵↵↵↵↵↵↵↵↵↵↵

1.23

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

1.23

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

1.22

Activations Density 0.762%