INDEX

Explanations

W followed by codes or numbers

The neuron strongly activates on isolated “w” (or “W”) tokens and shorthand tokens starting with that letter (e.g. “w” in an HTML ID or legal‐reporter abbreviations like “W.2d”).

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

 that

-1.63

ariConfig

-1.43

ícias

-1.38

 Giugno

-1.27

 évêque

-1.27

 Ejército

-1.27

</h4>

-1.24

sự

-1.24

ୟ

-1.22

 nestled

-1.21

POSITIVE LOGITS

culata

1.36

 algún

1.36

崤

1.35

BOTH

1.34

olkata

1.33

 SEVEN

1.30

 disminución

1.28

 alguma

1.28

 Seven

1.25

Baz

1.24

Activations Density 0.109%