INDEX

Explanations

pivotal states or qualities

The neuron fires strongly on rare, high-impact content words—especially dramatic or hyperbolic nouns and adjectives (e.g. “success,” “apocalypse,” “miracle,” “hell,” “nemesis,” “chameleon”).

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

to

-2.02

by

-1.93

if

-1.81

 from

-1.78

for

-1.78

 since

-1.69

 their

-1.64

of

-1.52

 Preußen

-1.48

 after

-1.45

POSITIVE LOGITS

がっ

1.69

わたし

1.63

㩒

1.61

教え

1.60

僦

1.59

気付

1.57

鷽

1.55

 vermelho

1.55

ちょこ

1.53

嗱

1.51

Activations Density 0.203%