INDEX

Explanations

commas separating list items

The neuron is detecting discourse-level transition markers (e.g. “In addition,” “Meanwhile,” “More importantly,” “Secondly,”).

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

了一会儿

-1.45

ensibility

-1.39

みたら

-1.36

のひとつ

-1.34

因为他

-1.34

nelse

-1.30

の一つ

-1.25

 Comprometido

-1.24

 pidiendo

-1.23

новые

-1.23

POSITIVE LOGITS

 then

1.66

1.57

1.51

.”

1.41

our

1.33

 further

1.32

<eos>

1.30

1.25

1.22

 additions

1.19

Activations Density 0.062%