INDEX

Explanations

sentence structure and phrasing

This neuron fires strongly on infrequent or “surprising” tokens—i.e. it marks out rare or unusual words (often sentence-initial or out-of-vocabulary terms).

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

 مجلس

0.57

 مردم

0.57

rägen

0.57

 realise

0.54

entretien

0.54

 Evaluations

0.54

 Gestion

0.53

ворю

0.53

Drs

0.53

 Employers

0.52

POSITIVE LOGITS

cking

0.64

washed

0.62

legged

0.61

 whopping

0.59

 mommy

0.59

footed

0.58

咦

0.58

!");

0.58

 yummy

0.58

édia

0.57

Activations Density 0.081%