INDEX

Explanations

communication in conflict

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

 Arctic

-0.09

ikel

-0.09

 reli

-0.09

unan

-0.09

 Answer

-0.09

nave

-0.08

 friendly

-0.08

Wake

-0.08

 Indo

-0.08

POSITIVE LOGITS

 Communication

0.13

 tone

0.12

Communication

0.12

 communication

0.12

tone

0.11

cal

0.11

 conflict

0.11

 assert

0.11

 hurt

0.11

æ²Ł

0.11

Activations Density 0.098%