INDEX

Explanations

CUP or Upgrade

The neuron is specialized to pick out tokens that are in uppercase or title‐case—especially acronyms (e.g. “CUP”) and metadata/header words (e.g. “Upgrade,” “Last,” “Version”).

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

肫

-2.75

 humeur

-2.59

悒

-2.58

猡

-2.56

梛

-2.55

 rosse

-2.53

-2.52

椟

-2.50

⸩

-2.48

 toalla

-2.47

POSITIVE LOGITS

2.58

2.52

2.50

 hohe

2.38

ではまた

2.34

}$.

2.33

 underlines

2.31

',

2.30

 cites

2.28

 anticipates

2.25

Activations Density 0.001%