INDEX

Explanations

explanation techniques

The neuron is detecting structural formatting tokens—especially punctuation marks (periods, question marks) and numbered/list-item markers.

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

ان

1.32

ل

1.29

ל

1.21

ر

1.15

ר

1.05

ا

1.02

ल

1.00

0.98

et

0.96

in

0.96

POSITIVE LOGITS

↵↵↵↵↵↵↵↵↵↵↵

1.02

↵↵↵↵↵↵↵↵

0.98

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.98

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.97

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.95

↵↵↵↵↵↵↵↵↵↵↵↵

0.93

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.93

↵↵↵↵↵↵↵

0.91

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.89

↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.89

Activations Density 24.460%