INDEX

Explanations

negative situations or states

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

loh

-0.10

âĨĴâĨĴ

-0.09

 PROT

-0.09

okoj

-0.09

instein

-0.09

 plast

-0.09

deb

-0.09

lest

-0.08

nton

-0.08

 drowned

-0.08

POSITIVE LOGITS

 toast

0.14

 trapped

0.12

 expend

0.12

 alone

0.11

 Toast

0.11

 halluc

0.11

 safest

0.10

-toast

0.10

 trouble

0.10

toast

0.10

Activations Density 0.040%