INDEX

Explanations

problem ask context input question

The neuron fires strongly on isolated, capitalized “name‐like” tokens—e.g. product or place names—so it’s spotting proper nouns (brand/item/place identifiers).

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

 специфи

0.72

னையும்

0.72

 regra

0.68

 acquies

0.67

 interoper

0.66

 armazen

0.65

 domínio

0.64

 circumvent

0.64

 précisément

0.64

 expropri

0.63

POSITIVE LOGITS

etc

0.75

Untitled

0.69

https

0.66

http

0.65

など

0.65

ר

0.63

ת

0.63

פ

0.63

Ashley

0.62

Activations Density 0.115%