INDEX

Explanations

inappropriate misuse or illegality

The neuron tends to light up on semantically “heavy” content words—rare or informative nouns, verbs, and other tokens that carry key meaning in a sentence.

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

 sayesinde

0.70

の変化

0.70

acyjne

0.67

 quieter

0.67

 கரைய

0.67

pozy

0.66

აწილ

0.65

ström

0.65

 중심으로

0.65

 uitgebre

0.65

POSITIVE LOGITS

 inappropriate

1.84

 unnecessary

1.77

 unnecessarily

1.75

 unwarranted

1.73

 inappropri

1.73

 unjustified

1.65

 needlessly

1.54

 unreasonable

1.53

 immoral

1.53

 unethical

1.53

Activations Density 1.504%