INDEX

Explanations

instructional or informational content

The neuron fires strongly on high‐content noun tokens—chiefly topic or section–heading words (e.g. “information,” “appointment,” “process,” “session”) that mark new sections or key concepts.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 цього

0.43

 this

0.39

this

0.39

 цієї

0.37

 цьому

0.37

 that

0.37

 such

0.37

such

0.36

 які

0.35

that

0.35

POSITIVE LOGITS

 enables

0.56

 exemplifies

0.54

 позволит

0.54

 umožňuje

0.52

 helps

0.50

 corresponds

0.50

 illustrates

0.49

 elucid

0.49

 allows

0.48

 explains

0.48

Activations Density 0.155%