INDEX

Explanations

programmatically / performed

The neuron fires on semantically rich “content” tokens (uncommon, information‐bearing words) rather than on trivial function words.

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

鯪

-1.44

 that

-1.41

还不错

-1.38

いつも

-1.37

 Commencez

-1.37

ｂ

-1.34

fillment

-1.33

まだ

-1.29

点数

-1.28

の場合は

-1.27

POSITIVE LOGITS

his

1.55

on

1.48

Vra

1.41

 your

1.34

 choć

1.26

 obé

1.26

 tránsito

1.25

 vieler

1.24

our

1.23

 young

1.22

Activations Density 0.340%