INDEX

Explanations

understated estimation or incorrectness

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

 misunderstood

-0.15

 misunderstand

-0.13

WithMany

-0.11

 disproportionate

-0.10

 neglected

-0.10

 misled

-0.10

 disproportionately

-0.09

 approximate

-0.09

;;;;;;

-0.08

-envelope

-0.08

POSITIVE LOGITS

 underst

0.27

 underestimate

0.26

 underestimated

0.23

 gross

0.21

est

0.18

gross

0.18

 Gross

0.16

 underrated

0.16

 Under

0.15

Est

0.14

Activations Density 0.055%