Is weak-to-strong generalization an alignment technique?
"可扩展监督" 相关文章
Balancing Label Quantity and Quality for Scalable Elicitation
How should we make trade-offs between the quantity and quality of labels used for eliciting knowledge from capable AI systems?
On scalable oversight with weak LLMs judging strong LLMs
Oversharing Details of NYU’s Work on Implementing Debate as an Alignment Technique
Scalable oversight as a quantitative rather than qualitative problem
用AI监督AI，OpenAI做到了用左脚踩右脚上天
GPT-4批评GPT-4实现「自我提升」，OpenAI前超级对齐团队又一力作被公开
