INDEX

Explanations

action by agent

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

Negative Logits

 innocent

-0.13

 terror

-0.10

 innoc

-0.09

 Threat

-0.09

 civilian

-0.09

Ã¤hr

-0.09

 threat

-0.09

 attempted

-0.09

Ø³Ø§Ø³

-0.09

threat

-0.08

POSITIVE LOGITS

 forces

0.18

 mistake

0.16

 superior

0.15

 hands

0.13

-products

0.12

 force

0.12

forces

0.11

 means

0.11

 gunfire

0.11

 jealous

0.11

Activations Density 0.100%

No Known Activations