INDEX

Explanations

"that's right"

The neuron fires strongly on semantically rich “content” tokens—key nouns, verbs, and proper nouns—rather than on high-frequency function words.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

but

-1.57

Rujuakan

-1.16

ͷ

-1.05

婚

-1.04

ственного

-0.96

ленным

-0.96

 qualquer

-0.95

 besonders

-0.91

)\,

-0.91

 אך

-0.90

POSITIVE LOGITS

 addirittura

1.22

 alltså

1.14

 nemmeno

1.10

就这么

1.05

 ufficiali

1.02

תוך

1.02

 altså

0.98

 basically

0.98

仅仅

0.93

Activations Density 0.044%