INDEX

Explanations

URLs and code comments

The neuron consistently activates on colon characters (both fullwidth and ASCII “:”).

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

 also

-1.67

 that

-1.45

 want

-1.38

 permitido

-1.34

 natomiast

-1.33

Unlike

-1.31

独特

-1.30

 both

-1.28

 primarily

-1.26

 therefore

-1.25

POSITIVE LOGITS

 https

1.34

ně

1.34

 ARISING

1.29

𝐒

1.29

 mereka

1.25

ändig

1.24

€“

1.20

https

1.17

ofas

1.17

AMERICAN

1.16

Activations Density 0.236%