INDEX

Explanations

This neuron fires strongly on non-prose tokens that mark webpage structure or navigation—things like bylines and author names, ellipses/section breaks, and “call-to-action” headings (e.g. KEEP READING HERE, Learn more).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ærk

0.51

 runner

0.50

 parametric

0.49

 designado

0.49

 (−

0.48

ravo

0.48

érin

0.47

 prossima

0.47

 वक्त

0.47

orsum

0.47

POSITIVE LOGITS

Read

1.09

全文

1.07

https

1.02

source

1.02

READ

1.00

 Read

0.98

read

0.98

 read

0.98

http

0.98

full

0.97

Activations Density 0.149%