INDEX

Explanations

medical compli

The neuron is strongly triggered by mid-word letter sequences—common roots or affixes (e.g. “reim,” “bur,” “pro,” “pos,” “uni,” “vers,” “come,” “ing,” “in”)—i.e. it detects internal morpheme fragments in longer words.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

॔

0.46

 soooo

0.42

 badass

0.39

 sooo

0.39

 skillet

0.36

 bijective

0.36

 fantastical

0.36

idk

0.35

ॕ

0.35

 tangy

0.35

POSITIVE LOGITS

Land

0.26

AI

0.26

重

0.26

越

0.26

Sport

0.25

Stef

0.25

三

0.25

前

0.25

将

0.25

om

0.25

Activations Density 0.140%