INDEX

Explanations

This blog

The neuron strongly activates on isolated short alphanumeric “codes” or identifiers—either sequences of digits (e.g. color‐code numbers) or uppercase letter pairs like “LC” and “HC.”

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

 even

-1.60

Even

-1.59

Although

-1.45

in

-1.44

The

-1.34

With

-1.34

-1.33

on

-1.32

 completely

-1.30

 Although

-1.27

POSITIVE LOGITS

Hoog

1.28

messungen

1.28

 navide

1.27

Initiative

1.26

Vri

1.25

𝟴

1.24

 ausein

1.23

𝟲

1.23

Cucumber

1.23

Etape

1.23

Activations Density 0.001%