INDEX

Explanations

create or structure

The neuron fires on content‐heavy nouns—especially multi-syllable, topic-specific nouns (e.g. “magazine,” “systems,” “callback,” “software,” “networks”).

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

 undertaken

-1.27

 implemented

-1.27

 being

-1.22

 chosen

-1.20

 established

-1.20

 developed

-1.20

 newly

-1.19

 devised

-1.19

 obtained

-1.17

 introduced

-1.16

POSITIVE LOGITS

for

2.47

 that

2.13

out

1.66

 around

1.56

 with

1.54

by

1.52

to

1.50

 capable

1.40

 from

1.30

 для

1.30

Activations Density 0.290%