INDEX

Explanations

prefixes of compound words and terms

The neuron fires on short technical tokens—especially single letters or brief uppercase abbreviations (e.g. “d,” “DK,” “ex,” “Full,” “per”) that often appear as parameter names, acronyms, or protocol keywords.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

It

-1.75

 bonté

-1.48

 récente

-1.48

дово

-1.44

 rękaw

-1.44

 deseado

-1.44

two

-1.40

The

-1.39

忘れて

-1.38

sklä

-1.38

POSITIVE LOGITS

 then

1.86

 such

1.60

纐

1.52

</sup>

1.41

 větší

1.38

,‎

1.34

then

1.31

 podían

1.31

and

1.30

 help

1.30

Activations Density 0.180%