INDEX

Explanations

numbers followed by `/`

The neuron selectively fires on the small numeric token (especially the “2” immediately before a “/”) that marks the start of a journal article’s DOI or identifier in reference 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

rawdę

-1.55

:'#

-1.48

.”

-1.48

 obtiene

-1.47

 使っ

-1.42

umerable

-1.41

iocene

-1.40

izop

-1.40

であった

-1.38

 العين

-1.36

POSITIVE LOGITS

驁

1.71

 ſte

1.71

proses

1.61

 consulter

1.59

橚

1.58

葙

1.55

Tahun

1.52

Muito

1.51

༈

1.49

敺

1.48

Activations Density 0.002%