INDEX

Explanations

before actions

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

lew

-0.10

oki

-0.10

 Bras

-0.09

 restarting

-0.09

 Ending

-0.09

Å

-0.09

 Clay

-0.08

 Shir

-0.08

peq

-0.08

IDS

-0.08

POSITIVE LOGITS

 proceeding

0.19

 Proceed

0.19

hand

0.17

-hand

0.17

 proceed

0.17

HAND

0.16

 proced

0.15

 proceeded

0.15

Proceed

0.15

 attempting

0.14

Activations Density 0.019%