INDEX

Explanations

website domain suffixes

The neuron activates on parts of web addresses—specifically top‐level domain tokens like “com,” “org,” “gov,” etc.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 http

0.73

http

0.68

www

0.63

WWW

0.62

 https

0.60

://

0.57

https

0.57

Christian

0.57

Wikipedia

0.55

LinkedIn

0.54

POSITIVE LOGITS

 Вайлдберриз

0.62

ത്സ

0.59

𝙩

0.55

 neutrality

0.54

 resorting

0.54

 sweets

0.54

 ஆனது

0.54

pisah

0.53

 обеспечи

0.52

लभ

0.52

Activations Density 0.096%