INDEX

Explanations

version of

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

ä¸Ģç§į

-0.13

ä¸ĢäºĽ

-0.12

ä¸ĢåĢĭ

-0.12

ä¸Ģ

-0.11

ä¸ĢçĤ¹

-0.11

ä¸ĢåĪĩ

-0.10

ä¸Ģäºº

-0.10

ä¸Ģçº§

-0.10

çļĦæĺ¯

-0.10

foon

-0.09

POSITIVE LOGITS

 sorts

0.12

 same

0.11

 reality

0.11

 otherwise

0.10

/cop

0.10

 itself

0.09

same

0.09

 Same

0.09

 truth

0.09

Reality

0.09

Activations Density 0.091%