INDEX

Explanations

misinformation and falsehoods

New Auto-Interp

Top Features by Cosine Similarity

Configuration

Prompts (Dashboard)

10,000 prompts, 128 tokens each

Dataset (Dashboard)

lmsys/lmsys-chat-1m

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

 misplaced

-0.11

TestFixture

-0.10

errupt

-0.10

 improperly

-0.10

åĪĴ

-0.09

teg

-0.09

~-~-

-0.09

 cach

-0.09

abet

-0.09

imei

-0.09

POSITIVE LOGITS

 false

0.25

false

0.21

 accuracy

0.18

 lies

0.18

 False

0.17

 errors

0.16

 fake

0.16

 accurate

0.16

False

0.15

 incorrect

0.15

Activations Density 0.139%