INDEX

Explanations

wrecked cars. Orangutan mother

The neuron fires strongly on bits of photo‐caption text—especially parenthetical color and contextual labels (e.g. “(in white)”, “(in orange)”) describing images.

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

 الفيلم

-1.30

ientras

-1.30

whom

-1.28

⁣⁣

-1.26

having

-1.25

珰

-1.23

랏

-1.23

 този

-1.23

ʹ

-1.23

 SEVEN

-1.21

POSITIVE LOGITS

Sometimes

1.31

 Sometimes

1.24

......

1.23

.......

1.21

all

1.21

These

1.19

 helped

1.17

eriks

1.16

Some

1.16

…..

1.16

Activations Density 0.250%