INDEX

Explanations

general information or context

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

oj

-0.10

 thoughtful

-0.09

 pledged

-0.09

isine

-0.09

 Shelter

-0.09

CHA

-0.09

eli

-0.08

 successful

-0.08

 undisclosed

-0.08

å·

-0.08

POSITIVE LOGITS

 ethical

0.19

 legitimate

0.17

Leg

0.15

ethical

0.15

 Ð±ÐµÐ·Ð¾Ð¿Ð°Ñģ

0.14

 ethics

0.13

 instead

0.13

Eth

0.13

 safer

0.13

LEG

0.13

Activations Density 0.175%