INDEX

Explanations

online publication or training

The neuron is detecting header‐style or metadata tokens—words or acronyms in all caps or title case (often appearing in report titles, locations, dates, and other heading elements).

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

 without

-1.66

 because

-1.63

 from

-1.59

 people

-1.51

not

-1.49

 everyone

-1.41

 especially

-1.34

 there

-1.29

by

-1.27

so

-1.26

POSITIVE LOGITS

まれた

1.37

 विश्लेषण

1.32

 একটি

1.30

agé

1.28

ècle

1.27

étrie

1.26

 Somit

1.21



1.20

 quoque

1.18

igshid

1.17

Activations Density 0.005%