cs.AI（2025-09-09）

📊 共 7 篇论文 | 🔗 2 篇有代码

🎯 兴趣领域导航

#	题目	一句话要点	标签	🔗
1	Certainty-Guided Reasoning in Large Language Models: A Dynamic Thinking Budget Approach	提出确定性引导推理(CGR)，提升大语言模型推理效率与可靠性。	large language model
2	XML Prompting as Grammar-Constrained Interaction: Fixed-Point Semantics, Convergence Guarantees, and Human-AI Protocols	提出基于XML提示的语法约束交互框架，保障LLM输出结构化和可控性。	large language model
3	Explicit Reasoning Makes Better Judges: A Systematic Study on Accuracy, Efficiency, and Robustness	研究表明，在LLM评判任务中，显式推理模型在准确性、效率和鲁棒性上更优。	large language model
4	VStyle: A Benchmark for Voice Style Adaptation with Spoken Instructions	VStyle：一个基于口语指令的语音风格迁移评测基准	instruction following	✅
5	Astra: A Multi-Agent System for GPU Kernel Performance Optimization	Astra：基于多智能体系统的GPU Kernel性能优化方法	large language model	✅

#	题目	一句话要点	标签	🔗	⭐
6	Language Self-Play For Data-Free Training	提出语言自博弈（LSP）方法，实现大模型在无数据条件下的持续改进。	reinforcement learning large language model instruction following
7	Competitive Audio-Language Models with Data-Efficient Single-Stage Training on Public Data	Falcon3-Audio：基于少量公共数据高效训练的竞争性音频-语言模型	curriculum learning large language model