INDEX

Explanations

rather than

The neuron responds to explicit contrast markers—particularly the words “not” and “rather” used to set up a comparison or negation.

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

 Dinas

-1.39

 hugely

-1.35

 quoique

-1.32

")){

-1.29

"*

-1.28

 אף

-1.27

 autant

-1.27

mê

-1.26

ar

-1.26

"/>

-1.25

POSITIVE LOGITS

to

1.70

 some

1.59

 only

1.47

 ONLY

1.45

 with

1.44

 just

1.32

 from

1.30

 because

1.30

 затем

1.29

 serves

1.24

Activations Density 0.064%