INDEX

Explanations

Results:

This neuron responds to tokens that are part of structured data elements—especially numeric values and uppercase identifiers (e.g. table entries, infobox numbers, acronyms).

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

 상세

-1.41

endforeach

-1.33

aic

-1.32

abstractmethod

-1.32

лася

-1.31

 Belgische

-1.28

Ņ

-1.27

∣

-1.26

 oublié

-1.25

 mogli

-1.25

POSITIVE LOGITS

 with

1.91

any

1.78

how

1.77

 after

1.38

 their

1.38

 evitare

1.34

 дать

1.31

 оставить

1.29

 even

1.28

 current

1.25

Activations Density 0.969%