INDEX

Explanations

The neuron fires on specific descriptive or qualifying words—especially adjectives, quantifiers, or proper names—that add concrete detail (e.g. “internal,” “few,” “Turkish,” “Leslie,” “many,” “short,” “spinal,” “match,” “provincial,” “district”).

short notice

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

to

-1.34

<bos>

-1.30

參考文獻

-1.20

時間は

-1.16

of

-1.14

 with

-1.09

now

-1.08

یزات

-1.03

is

-0.99

時間が

-0.97

POSITIVE LOGITS

 behalf

2.36

 occasion

1.28

corrhi

1.23

 hohem

1.23

ifax

1.20

steroids

1.16

 ועל

1.15

 nuits

1.14

 steroids

1.13

為に

1.13

Activations Density 0.104%