INDEX

Explanations

title followed by subject

The neuron fires on bits of programming source (single-letter identifiers like “b” or “v,” comment‐marker tokens like “**,” filenames/extensions, and code keywords), i.e. it detects code‐style tokens.

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

 then

-1.60

 they

-1.57

 çünkü

-1.49

groet

-1.45

 financiación

-1.31

 étages

-1.30

 этой

-1.28

瘩

-1.26

 tell

-1.25

VStack

-1.25

POSITIVE LOGITS

but

1.77

 However

1.58

 tumbuhan

1.52

有不少

1.42

órica

1.42

TWO

1.41

listair

1.40

комплект

1.40

趼

1.39

Although

1.38

Activations Density 0.000%