INDEX

Explanations

list items and code blocks

The neuron fires most strongly on short uppercase abbreviations—i.e. acronyms or initialisms (whether internet slang like LOL/IMHO or code macros like PRAGMA/ONCE).

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

 from

-1.38

)=$

-1.23

}$\

-1.20

)+

-1.19

ксана

-1.14

එ

-1.13

bruik

-1.13

økt

-1.12

âmica

-1.11

 each

-1.10

POSITIVE LOGITS

or

1.65

 they

1.39

 progetto

1.26

彼らは

1.20

 compleanno

1.15

 bugün

1.15

or

1.14

拿出

1.13

catálogo

1.12

 nothing

1.11

Activations Density 0.532%