INDEX

Explanations

The neuron fires strongly on capitalized multi‐word phrases or headings—i.e. title-style proper names and section titles.

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

 atleast

0.67

datatype

0.66

 እየሱስ

0.66

reversed

0.63

макраты

0.62

égalité

0.62

 expts

0.61

 بهم

0.60

fillna

0.59

 upto

0.59

POSITIVE LOGITS

↵↵

1.61

1.59

↵↵↵↵

1.57

↵

1.52

↵↵↵

1.47

<start_of_image>

1.40

↵↵↵↵↵

1.36

｜

1.31

↵↵↵↵↵↵

1.30

 Although

1.20

Activations Density 1.213%