INDEX

Explanations

actions and their objectives

The neuron activates on content-heavy action verbs (especially less common or domain-specific ones, like “searching,” “compiled,” “tests,” “creates,” “signed,” etc.).

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

in

-1.06

 just

-0.98

so

-0.95

唁

-0.91

did

-0.91

or

-0.89

gı

-0.86

 after

-0.86

 時代

-0.86

not

-0.85

POSITIVE LOGITS

to

1.33

 perumahan

1.16

を入れる

1.02

 têm

0.93

 DRAFT

0.91

 !!!!

0.91

 !!!!!

0.90

歟

0.89

 жаль

0.89

RoleId

0.89

Activations Density 0.087%