INDEX

Explanations

Sure, as, 1, here, to, executive

assistant-style preamble and formatting at the start of responses, including polite openings, role/self-identification, and list introductions, with strong sensitivity to line breaks and commas.

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

Negative Logits

elp

-0.11

afil

-0.10

rr

-0.09

.ix

-0.09

ess

-0.08

boxed

-0.08

otor

-0.08

 Maced

-0.08

ivia

-0.08

 Beau

-0.08

POSITIVE LOGITS

ä¸ºäºĨ

0.15

 Based

0.15

 based

0.14

Based

0.13

 Below

0.12

éĴĪ

0.12

Below

0.12

ä»¥ä¸ĭ

0.12

 lets

0.11

 below

0.11

Activations Density 0.280%