INDEX

Explanations

victims, survivors, recipients

The neuron strongly activates on prominent, content‐bearing tokens—often capitalized words or numbers—that introduce new sections or serve as headings (e.g. “legal,” “Powder,” “Holocaust,” “Film,” “7”) rather than on ordinary 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

Negative Logits

经纪

0.71

经理

0.63

 प्रतिस्पर्धा

0.60

医師

0.59

 руководством

0.57

不利

0.56

 brokerage

0.55

 менедже

0.55

 род

0.54

 pulsation

0.54

POSITIVE LOGITS

 winners

1.93

 winner

1.89

winners

1.66

 survivors

1.64

 recipients

1.62

winner

1.62

 losers

1.57

Winner

1.47

 victims

1.46

 recipient

1.45

Activations Density 0.628%