INDEX

Explanations

apologies and negations

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

 seriously

-0.10

lis

-0.10

ill

-0.10

 thank

-0.10

 pall

-0.09

-Sah

-0.09

und

-0.09

ful

-0.08

AspNet

-0.08

yes

-0.08

POSITIVE LOGITS

 sorry

0.35

Sorry

0.31

 Sorry

0.31

sorry

0.30

 tiáº¿

0.29

éģĹ

0.27

Unfortunately

0.26

 regret

0.26

SOR

0.24

æĬ±

0.24

Activations Density 0.322%