研究人员开发了“optimize_anything”，一个通用API，通过将各种优化问题视为基于文本的改进来解决它们。该系统在各种任务中展示了最先进的结果，包括增强AI代理架构、优化云调度算法和生成高效的CUDA内核。研究强调，提供可操作的侧面信息和采用多任务学习，与仅基于分数的反馈或独立优化相比，可以显著提高收敛速度和最终得分。 AI
影响 这种新的优化范式可以将各种问题求解任务统一到一个基于LLM的框架下，有可能简化开发并在不同领域提高性能。
排序理由 该集群包含一篇详细介绍基于LLM的优化新方法的学术论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
Can a single LLM-based optimization system match specialized tools across fundamentally different domains? We show that when optimization problems are formulated as improving a text artifact evaluated by a scoring function, a single AI-based optimization system-supporting single-…
Can a single LLM-based optimization system match specialized tools across fundamentally different domains? We show that when optimization problems are formulated as improving a text artifact evaluated by a scoring function, a single AI-based optimization system-supporting single-…