INDEX

Explanations

code and license headers

The neuron tends to fire on rare or unusually specific tokens—long identifiers, proper names, version‐number literals, or other low-frequency words—rather than common function words.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

hofer

-0.83

 että

-0.75

生存

-0.71



-0.71

被告

-0.71

rikan

-0.71

ventura

-0.71

檻

-0.70

Callable

-0.70

 additional

-0.70

POSITIVE LOGITS

XT

0.83

Cambio

0.79

Hint

0.77

hame

0.75

빙

0.75

ⓑ

0.75

مثل

0.74

聖誕

0.74

▏

0.74

 Vitro

0.73

Activations Density 0.003%