INDEX

Explanations

retrospectively

The neuron fires on high-level section headings or structural markers (e.g. “Abstract:”, “Description of the Prior Art:”, “Specific Aim”, numbered or titled document sections).

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

-1.46

 these

-1.41

もありました

-1.35

 those

-1.21

中でも

-1.15

may

-1.13

などは

-1.13

でも

-1.13

 especially

-1.11

at

-1.09

POSITIVE LOGITS

様な

1.42

様に

1.28

 finos

1.18

くんの

1.17

にします

1.16

為に

1.13

</em>

1.11

皷

1.09

感じで

1.09

 Temos

1.09

Activations Density 0.019%