INDEX

Explanations

prohibitions and obligations

The neuron fires on emphatic, all-caps warning or instruction tokens (e.g. WARNING, NOT, MUST, YOU WILL) and similar high-urgency markers.

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

 with

-1.80

 being

-1.41

of

-1.37

 straff

-1.34

 helping

-1.31

 gradually

-1.29

 found

-1.28

 добраться

-1.25

 ideias

-1.25

 visión

-1.25

POSITIVE LOGITS

 however

1.39

 especializada

1.30

 отсутствие

1.28

这

1.26

however

1.24

 nová

1.23

 появление

1.23

&&

1.21

therefore

1.20

 toutefois

1.20

Activations Density 0.081%