INDEX

Explanations

This neuron fires on domain-specific content words—especially technical or scientific nouns like “performance,” “chemical,” or “operation.”

the performance analysis

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

to

-2.17

but

-1.20

 being

-1.15

 subsequent

-1.09

by

-1.09

⩽

-1.05

 inability

-1.01

 able

-1.00

肇

-0.98

 other

-0.98

POSITIVE LOGITS

というわけで

1.57

＊＊＊

1.31

こちらが

1.30

＊＊

1.30

そういえば

1.29

そりゃ

1.28

 penna

1.25

うーん

1.23

引用元

1.20

tells

1.20

Activations Density 0.062%