INDEX

Explanations

code comment and headers

This neuron fires strongly on document‐structure markers—especially markdown‐style headings (lines beginning with “Q:”, “###”, “##”, etc.).

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

-1.82

-1.73

我們

-1.56

 такой

-1.54

曾经

-1.52

曾經

-1.45

我们

-1.45

 вашего

-1.44

we

-1.42

 hinzugefügt

-1.41

POSITIVE LOGITS

</h2>

1.58

1.46

◆◆

1.46

⚫

1.41

</sup>

1.40

👫

1.39

特定的

1.36

 specifically

1.36

这是什么

1.33

ise

1.31

Activations Density 0.181%