INDEX

Explanations

square brackets around assistant responses

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

 twin

-0.10

-0.09

iado

-0.09

"urls

-0.08

ãģĭãĤı

-0.08

 -*-č\n

-0.08

ÃĸL

-0.08

marshall

-0.08

Ã¼f

-0.08

'''č\n

-0.08

POSITIVE LOGITS

`[

0.19

 brackets

0.17

 bracket

0.16

0.15

 square

0.15

0.14

'['

0.14

Bracket

0.13

 Square

0.12

\[

0.12

Activations Density 0.093%