INDEX

Explanations

special characters and code formatting

The neuron strongly activates on numeric literal tokens (e.g. “126,” “152,” etc.) in structured data like XML, i.e. it detects standalone numbers.

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

of

-1.69

to

-1.38

get

-1.33

 found

-1.30

 their

-1.29

if

-1.26

all

-1.23

 from

-1.20

not

-1.19

at

-1.16

POSITIVE LOGITS

),

1.28

quelize

1.23

Lastname

1.23

鉏

1.21

镂

1.17

strzy

1.16

чної

1.16

 "..\..\..\

1.13

お知らせ

1.13

這裡

1.13

Activations Density 0.001%