Learnable Behavior Control: Breaking Atari Human World Records via Sample-Efficient Behavior Selection
"行为控制" 相关文章
Prompt Decorators: A Declarative and Composable Syntax for Reasoning, Formatting, and Control in LLMs
Prompt Decorators: A Declarative and Composable Syntax for Reasoning, Formatting, and Control in LLMs
Prompt Decorators: A Declarative and Composable Syntax for Reasoning, Formatting, and Control in LLMs
VISOR++: Universal Visual Inputs based Steering for Large Vision Language Models
给大模型「精准手术」：美团智能客服提出逆向学习技术精准纠偏，风险控制提升38%
Psychometric Personality Shaping Modulates Capabilities and Safety in Language Models
VISOR: Visual Input-based Steering for Output Redirection in Vision-Language Models
Forcing LLMs to be evil during training can make them nicer in the long run
Comparing the effectiveness of top-down and bottom-up activation steering for bypassing refusal on harmful prompts
What You (Want to)* Want
