INDEX

Explanations

design & structure

This neuron detects publication-metadata cues (e.g. “appeared,” “first,” “Read more,” newswire labels) that mark bylines and source information.

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

 integral

0.23

 ancillary

0.23

 palpable

0.23

 induc

0.23

 superlative

0.23

 intermitt

0.23

 impor

0.23

 attent

0.22

 indisc

0.22

 mele

0.22

POSITIVE LOGITS

Э

0.32

ر

0.30

 Фор

0.29

 Фа

0.29

وب

0.28

İ

0.27

Э

0.27

Фа

0.27

 Ху

0.27

Activations Density 0.071%