PRIVATELY ALIGNING LANGUAGE MODELS WITH REINFORCEMENT LEARNING

Keyphrases

• Language Model 100%
• Reinforcement Learning 100%
• Differential Privacy 40%
• Large Language Models 40%
• Instruction Following 20%
• Human-in-the-loop 20%
• Privacy-preserving 20%
• Dominant Paradigm 20%
• Following Model 20%
• Privacy Protection 20%
• Positive Reviews 20%
• Privacy Framework 20%
• Training Instruction 20%
• Review Generation 20%
• Chat Generative Pre-trained Transformer (ChatGPT) 20%
• Reinforcement Learning from Human Feedback 20%

Computer Science

• Reinforcement Learning 100%
• Language Modeling 100%
• Differential Privacy 33%
• Large Language Model 33%
• Experimental Result 16%
• Privacy Preserving 16%
• Dominant Paradigm 16%
• Privacy Protection 16%
• Human-in-the-Loop 16%
• ChatGPT 16%