INDEX

Explanations

supporting corruption

The neuron fires strongly on mentions of large‐scale monetary figures or financial wrongdoing, i.e. references to sizeable sums or fraud.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

onday

0.68

 Timeout

0.65

SeekBar

0.64

挑戰

0.64

开口

0.64

攻

0.63

 उत्पा

0.62

ubuntu

0.61

Attack

0.61

сног

0.61

POSITIVE LOGITS

 fraudulently

1.77

 fraudulent

1.71

 fraud

1.67

 embezz

1.66

 embezzlement

1.66

 defraud

1.62

 corrupt

1.56

 defra

1.52

fraud

1.50

 frauds

1.45

Activations Density 0.613%