INDEX

Explanations

purposes and clarity

This neuron fires strongly on “for … with …” template phrases—i.e. descriptive or promotional fragments of the form “for [adjective] [noun] with [adjective] [noun].”

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

 для

1.80

 für

1.66

 για

1.62

for

1.59

 voor

1.55

 pentru

1.55

 untuk

1.54

 برای

1.52

для

1.52

for

1.47

POSITIVE LOGITS

 sake

1.96

 purposes

1.70

来说

1.69

 Sake

1.54

來說

1.44

 Purposes

1.40

而言

1.23

来讲

1.06

 purpose

0.86

 целей

0.81

Activations Density 0.707%