INDEX

Explanations

father did, extremely protective

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

 victims

-0.10

ç¾

-0.09

reetings

-0.09

arah

-0.09

 absent

-0.09

.ld

-0.09

LD

-0.08

Backing

-0.08

ADR

-0.08

igne

-0.08

POSITIVE LOGITS

 incons

0.14

 shock

0.14

 breakdown

0.14

 grief

0.13

 Shock

0.13

Shock

0.12

 Lear

0.12

 devastated

0.12

æĤ²

0.12

 bedside

0.11

Activations Density 0.056%