INDEX

Explanations

he had / the direction

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

arin

-0.10

rena

-0.10

undo

-0.09

ois

-0.09

 Barton

-0.09

ilia

-0.09

 Sund

-0.09

bye

-0.08

 Cory

-0.08

POSITIVE LOGITS

 danger

0.16

 attention

0.15

 Attention

0.14

 approaching

0.13

 presence

0.13

 dangers

0.12

 Appro

0.12

 upcoming

0.12

attention

0.12

Appro

0.12

Activations Density 0.086%