INDEX

Explanations

The neuron fires most strongly on uncommon, content-bearing tokens (i.e. low-frequency, information-rich words rather than common function words).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

这样一个

0.87

 предприя

0.82

 जिसने

0.82

 заявление

0.82

 prestataire

0.80

istage

0.78

创建一个

0.78

 seorang

0.77

 operativo

0.77

проце

0.77

POSITIVE LOGITS

 각각

1.34

 যেগুলো

1.30

 vertices

1.27

 groupings

1.26

 amelyek

1.26

 names

1.25

 jeweils

1.22

 grouped

1.21

 nomes

1.21

 pairs

1.20

Activations Density 3.319%