INDEX

Explanations

attempts to jailbreak or manipulate the AI into violating its guidelines and generating inappropriate content.

oai_token-act-pair · claude-4-5-haiku Triggered by @xiaoqingsun004

New Auto-Interp

Configuration

andyrdt/saes-qwen2.5-7b-instruct/resid_post_layer_19/trainer_1

Dataset (Dashboard)

Various

Features

131,072

Data Type

float32

Hook Name

blocks.19.hook_resid_post

Architecture

standard

Context Size

1,024

Dataset

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

енные

-0.08

↜

-0.07

ivas

-0.07

宮

-0.07

GW

-0.07

ﰢ

-0.07

 Copyright

-0.07

={"/

-0.07

 describe

-0.06

corp

-0.06

POSITIVE LOGITS

 свет

0.09

青睐

0.07

GLenum

0.07

 büyü

0.07

乌鲁

0.07

手势

0.07

 bụng

0.07

луч

0.07

 bekannt

0.07

 recycle

0.07

Activations Density 0.647%

attempts to jailbreak or manipulate the AI into violating its guidelines and generating inappropriate content.

No Comments

No Known Activations