INDEX

Explanations

The participants examination majority study selection autoregressive case

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

 mist

-0.09

 pert

-0.09

aty

-0.09

å±

-0.09

aga

-0.08

 ding

-0.08

 rehears

-0.08

 Ding

-0.08

 presum

-0.08

::$_

-0.08

POSITIVE LOGITS

 latter

0.12

oretical

0.11

udit

0.10

integral

0.10

irs

0.09

amil

0.08

GET

0.08

zia

0.08

 authors

0.08

Activations Density 0.084%