INDEX

Explanations

This neuron activates on formal academic‐style metadata and section headings—e.g. acknowledgments, calls for participation, conference names or dates, and other high-level paper info.

conference proceedings and papers

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

 ویتامین

-1.77

 ингреди

-1.62

 витами

-1.48

my

-1.41

notag

-1.38

 сахара

-1.34

-1.29

 when

-1.24

 окружающей

-1.23

-1.21

POSITIVE LOGITS

וּ

1.90

וֹ

1.41

 Bewußt

1.36

־ה

1.35

 discípulos

1.35

fikation

1.34

 IEEE

1.33

 šaty

1.33

 JMenuItem

1.31

ſo

1.29

Activations Density 0.016%