INDEX

Explanations

Good, work, Porsche, Cloud names

This neuron fires strongly on section headers and bibliographic references—specifically book titles and author names (often rendered in uppercase or set off as a quote).

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

 receipts

-0.86

 спек

-0.81

itemCount

-0.77

agreb

-0.77

流畅

-0.75

çait

-0.74

)$

-0.73

imt

-0.73

끗

-0.72

Carb

-0.72

POSITIVE LOGITS

 Southwest

0.81

 Hedgehog

0.75

ಊ

0.74

 Senioren

0.74

hr

0.73

 replace

0.71

+='

0.71

リム

0.70

ُ

0.69

)^{-

0.69

Activations Density 0.019%