INDEX

Explanations

can we

The neuron fires most strongly on modal‐of‐possibility words and phrases (e.g. “can,” “possible,” “whether”) that signal ability or likelihood, especially in question contexts.

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

 either

-1.80

Tujuan

-1.52

Cinta

-1.52

fashionable

-1.52

Sebagai

-1.51

realme

-1.51

replyreport

-1.50

 femininas

-1.49

 teile

-1.48

【

-1.48

POSITIVE LOGITS

an

1.72

of

1.61

1.49

1.48

be

1.45

 https

1.45

1.43

1.41

他的

1.40

Activations Density 0.089%