INDEX

Explanations

sequence of actions

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

 proceeding

-0.09

 facet

-0.09

 Aaron

-0.09

 reserved

-0.08

 approach

-0.08

 Incoming

-0.08

 Warn

-0.08

 democrat

-0.08

 Chatt

-0.08

aves

-0.08

POSITIVE LOGITS

 actions

0.17

 Actions

0.15

 movement

0.13

 sequence

0.12

actions

0.12

-actions

0.12

 movements

0.11

_actions

0.11

Actions

0.11

sequence

0.11

Activations Density 0.098%