INDEX

Explanations

may not be

The neuron fires on concessive negation constructions – essentially the “not … but” pattern (e.g. “may not know it, but …”).

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

 ソファ

-1.48

 Что

-1.41

 якщо

-1.40

 demandé

-1.39

itado

-1.34

 rencontré

-1.32

⁫

-1.32

 essayé

-1.32

Bagaimana

-1.30

Kelebihan

-1.29

POSITIVE LOGITS

 then

2.00

 nawet

1.56

 even

1.54

 your

1.52

 later

1.48

 особенно

1.48

 following

1.36

 permiten

1.36

、「

1.32

 necessarily

1.29

Activations Density 0.052%