INDEX

Explanations

code definitions and examples

The neuron activates on code‐style tokens—particularly underscores and snake_case identifiers (e.g. method or variable names).

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

 more

-1.39

 both

-1.34

 only

-1.33

 first

-1.30

for

-1.28

 some

-1.27

 very

-1.26

 Although

-1.23

 would

-1.20

 other

-1.20

POSITIVE LOGITS

udert

1.38

 praticamente

1.37

 ayur

1.36

 maniere

1.34

 certamente

1.34

 niedersachsen

1.32

쫑

1.31

摀

1.30

킁

1.27

癭

1.27

Activations Density 0.011%