INDEX

Explanations

Bad data, overflow, buffer

The neuron fires on programming identifiers and labels (e.g. variable names or test‐case markers like “Bad Source,” “Good Source,” “sources,” “data,” “u,” “i,” “le,” 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

or

-1.57

 them

-1.53

to

-1.52

 princí

-1.45

in

-1.36

 nejs

-1.35

 desmotiv

-1.34

 faciles

-1.34

as

-1.32

 from

-1.30

POSITIVE LOGITS

1.95

1.75

1.71

.............

1.59

这也是

1.56

—.

1.53

we

1.51

1.48

............

1.47

Activations Density 0.001%