INDEX

Explanations

distribution, is, annotations

The neuron flags words used as section‐heading or call‐to‐action tokens in a web page (e.g. “Weather,” “Download,” “push,” “photo,” “picture”).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 Sunni

0.63

istic

0.55

(;

0.55

无可

0.55

0.54

.(

0.53

 खार

0.52

 }^{(

0.51

ValMap

0.51

ammad

0.51

POSITIVE LOGITS

 wretched

0.60

 hlav

0.54

 ناخذ

0.54

unfinished

0.54

 hous

0.54

andelions

0.54

ْت

0.53

 dilap

0.53

ресенье

0.53

═

0.53

Activations Density 0.000%