INDEX

Explanations

questions and conversational starters

The neuron lights up on sentence‐initial, dialogue-style words—especially pronouns (“I,” “you,” “must,” “be”) and question/response starters (“What,” “Is,” “Do,” “Okay”).

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

どうしよう

-0.95

SetData

-0.93

нинград

-0.91

Personen

-0.88

COLN

-0.88

Fazit

-0.87

௧

-0.86

 somme

-0.85

 Sénat

-0.85

 certiorari

-0.85

POSITIVE LOGITS

 What

1.78

 what

1.73

How

1.56

how

1.52

What

1.51

but

1.47

 then

1.46

Why

1.38

why

1.34

so

1.30

Activations Density 0.017%