INDEX

Explanations

event awards

The neuron strongly activates on long or unusual tokens—whether technical identifiers, rare multi-syllable words, or standalone numbers—i.e. it’s looking for those long/rare “specialized” tokens.

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

uerto

-0.77

ensen

-0.73

PSF

-0.73

Yay

-0.69

 chisel

-0.65

erjee

-0.64

Rück

-0.63

 Эй

-0.62

 senso

-0.61

ắc

-0.61

POSITIVE LOGITS

 Vande

0.71

치

0.70

rekt

0.68

 denke

0.66

 Scene

0.66

 Ages

0.65

 Gele

0.65

ится

0.64

 legais

0.64

 instructional

0.63

Activations Density 0.086%