INDEX

Explanations

both sides

New Auto-Interp

Configuration

Prompts (Dashboard)

16,384 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

fea

-0.11

 deter

-0.10

bral

-0.10

jak

-0.10

anel

-0.10

 deepen

-0.10

entine

-0.10

 extras

-0.10

COND

-0.10

arak

-0.09

POSITIVE LOGITS

 directions

0.18

 sides

0.17

 sexes

0.13

 senses

0.12

 genders

0.11

 respects

0.11

 direction

0.11

 halves

0.11

heimer

0.09

Directions

0.09

Activations Density 0.054%

No Known Activations