强化学习AI系统的设计实现及未来发展
"RLHF" 相关文章
小互 💬 : AI 公司们为了获取“高质量”的语料，把大量 19 世纪末、20 世纪初的文学作品喂给了模型。 AI 忠实地学习了那个年代的写作风格——其中就包括对“破折号”的狂热喜爱。
Beyond Scale: Why RLHF Is the Future of Specialized AI
Beyond Scale: Why RLHF Is the Future of Specialized AI
Why do AI models use so many em-dashes?
Why do AI models use so many em-dashes?
What can we learn from parent-child-alignment for AI?
What can we learn from parent-child-alignment for AI?
Greedy Sampling Is Provably Efficient for RLHF
PaTaRM: Bridging Pairwise and Pointwise Signals via Preference-Aware Task-Adaptive Reward Modeling
When Personalization Meets Reality: A Multi-Faceted Analysis of Personalized Preference Learning
Anthropic、Thinking Machines Lab论文曝光：30万次压力测试揭示AI规范缺陷
一个小技巧，帮你显著提高 AI 的回答质量！
一个小技巧，帮你显著提高 AI 的回答质量！
LLM Training Data Optimization: Fine-Tuning, RLHF & Red Teaming
LLM Training Data Optimization: Fine-Tuning, RLHF & Red Teaming
Direct Preference Optimization with Unobserved Preference Heterogeneity: The Necessity of Ternary Preferences
GPT-5 核心成员详解 RL：Pre-training 只有和 RL 结合才能走向 AGI
GPT-5 核心成员详解 RL：Pre-training 只有和 RL 结合才能走向 AGI
GPT越来越保守？斯坦福Manning团队提出Verbalized Sampling，让模型重新“多想一点”
