INDEX

Explanations

system components or conditions

The neuron strongly activates on document‐style metadata words (e.g. “TOC,” “providers,” “Internet,” “security,” “ventilation”)—in other words, it picks out report/section‐heading labels and similar structural or technical terms.

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

지와

0.96

人和

0.94

и

0.91

and

0.91

 এবং

0.89

 και

0.89

และ

0.89

and

0.88

 और

0.86

 અને

0.85

POSITIVE LOGITS

 sebagainya

0.88

 ஆகியவை

0.87

 ஆகியவற்ற

0.76

 देखील

0.75

 있었습니다

0.71

0.68

 insinu

0.68

 posiblemente

0.68

。【

0.67

 উভয়

0.66

Activations Density 0.001%