INDEX

Explanations

question labels

This neuron strongly activates on metadata tokens from generated code or config files—e.g. XML tag names (like “list”) and autogenerated file identifiers (like “82_good_G2B86”). In other words, it flags boilerplate or structural markers from code/config headers.

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

in

-1.80

 their

-1.66

 this

-1.47

 from

-1.44

of

-1.42

by

-1.37

 high

-1.34

any

-1.33



-1.31

 encanto

-1.29

POSITIVE LOGITS

 alcuni

1.59



1.31

 astonishing

1.30

 alguns

1.30

 amazingly

1.25

 insanely

1.25

 alcune

1.24

 diversos

1.24

mainly

1.24

 dilapidated

1.23

Activations Density 0.000%