INDEX

Explanations

apology + but + inability

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

 Chall

-0.10

ernet

-0.09

æ¸Ī

-0.09

ior

-0.08

ä¹Łæĺ¯

-0.08

uire

-0.08

 Feinstein

-0.08

ensburg

-0.08

odus

-0.08

_ALREADY

-0.08

POSITIVE LOGITS

ä¸įèĥ½

0.14

 unable

0.13

 cannot

0.13

æĹłæ³ķ

0.13

cannot

0.12

 couldn

0.11

Looks

0.11

 Cannot

0.11

 Looks

0.11

 Unable

0.10

Activations Density 0.111%