INDEX

Explanations

a short

This neuron isn’t really detecting a semantic pattern in the words but rather their position in the sequence—it fires on tokens that fall at a regular interval through the document, acting like a periodic positional marker.

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

 الا

0.85

 بم

0.84

มี

0.79

mak

0.77

func

0.75

txt

0.74

da

0.73

ca

0.73

ni

0.72

force

0.72

POSITIVE LOGITS

чества

0.88

 discontent

0.77

 disgr

0.74

 cercando

0.74

réd

0.73

 courtship

0.73

ून

0.72

 ceremonies

0.72

 piety

0.72

 chinos

0.72

Activations Density 0.000%