INDEX

Explanations

pig latin

The neuron fires almost exclusively on isolated short “label” tokens that head new sections—i.e. single‐word or few‐letter headings, abbreviations, or acronyms (R, J, N9, Pro, Crash, Rub, div, etc.) rather than normal running text.

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

-1.47

-1.43

ts

-1.34

部分が

-1.26

www

-1.26

de

-1.26

one

-1.24

es

-1.23

POSITIVE LOGITS

鹀

1.40

 Codable

1.37

swire

1.34

 légales

1.27

 toneladas

1.26

śred

1.26

цена

1.25

ቍ

1.25

leun

1.25

⸒

1.23

Activations Density 0.208%