INDEX

Explanations

<|start_header_id|>assistant<|end_header_id|>

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

 seedu

-0.10

Ã¼rn

-0.09

ighton

-0.09

ontent

-0.08

othy

-0.08

 Fehler

-0.08

 diffuse

-0.08

rzy

-0.08

amedi

-0.08

otime

-0.08

POSITIVE LOGITS

 handling

0.19

 Handling

0.16

 handle

0.15

 safety

0.14

 Handle

0.14

Handling

0.14

 handled

0.13

 Safety

0.13

MSD

0.12

mut

0.12

Activations Density 0.033%