INDEX

Explanations

dominance, Diff, CNT, morale

The neuron fires on rare or domain‐specific “jargon” tokens—things like technical shorthand (e.g. C++ header file extensions “.hh”), programming terms (“Diff”), biology terms (“dominance”), etc.—i.e. unusually uncommon or specialized content words.

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

-3.64

</em>

-2.75

錸

-2.70

to

-2.69

}),

-2.58

ा

-2.56

mindedness

-2.48

蛧

-2.45

-2.39

halation

-2.38

POSITIVE LOGITS

With

2.70

There

2.55

 gerçekten

2.47

They

2.47

These

2.45

 tweaked

2.44

After

2.42

This

2.42

依然

2.42

࿚

2.38

Activations Density 0.033%