让 Agent 更可靠地调用工具和复用技能、识别并缓解模型安全、越狱和对齐风险

今天最值得跟进的方向

今天的高分论文主要指向:让 Agent 更可靠地调用工具和复用技能、让 Agent 更可靠地调用工具和复用技能、识别并缓解模型安全、越狱和对齐风险。建议先看每篇的原文链接、摘要、评测设置和代码/数据是否可用,再决定是否深入复现。

重点论文:题目、看点与核验线索

1. 让 Agent 更可靠地调用工具和复用技能

Cosmos 3: Omnimodal World Models for Physical AI (Aditi, Niket Agarwal, Arslan Ali, Jon Allen, Martin Antolini, Adeline Aubame, et al.) 2606.02800 PDF

让 Agent 更可靠地调用工具和复用技能。摘要显示:We introduce Cosmos 3, a family of omnimodal world models designed to jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture. 重点核验:任务设置是否真实,是否有代码或数据,评测是否覆盖复杂场景,结论是否能迁移到实际系统。

2. 让 Agent 更可靠地调用工具和复用技能

Thinking Past the Answer: Evaluating Harmful Overthinking in Large Reasoning Models (Simone Caldarella, Davide Talon, Rahaf Aljundi, Elisa Ricci, Massimiliano Mancini) 2606.02835 PDF

让 Agent 更可靠地调用工具和复用技能。摘要显示:Large Reasoning Models (LRMs) improve performance by generating explicit intermediate reasoning traces through increased test-time compute, yet the assumption that longer reasoning is consistently beneficial remains under-examined. 重点核验:任务设置是否真实,是否有代码或数据,评测是否覆盖复杂场景,结论是否能迁移到实际系统。

3. 识别并缓解模型安全、越狱和对齐风险

Breaking the Information Silo: Semantic Personas for Cross-Domain Recommendation (Jonathan Mayo, Moshe Unger, Konstantin Bauman) 2606.01783 PDF

识别并缓解模型安全、越狱和对齐风险。摘要显示:Digital platforms increasingly operate as isolated information silos, limiting their ability to construct comprehensive user representations across domains. 重点核验:任务设置是否真实,是否有代码或数据,评测是否覆盖复杂场景,结论是否能迁移到实际系统。

4. 让 Agent 更可靠地调用工具和复用技能

KForge: LLM-Driven Cross-Platform Kernel Generation for AI Accelerators (Taras Sereda, Burak Bartan, Ankita Nayak, Tom St. John, Natalie Serrino, Zain Asgar) 2606.02963 PDF

让 Agent 更可靠地调用工具和复用技能。摘要显示:Production inference increasingly targets a heterogeneous mix of accelerators. 重点核验:任务设置是否真实,是否有代码或数据,评测是否覆盖复杂场景,结论是否能迁移到实际系统。

5. 提升代码生成、执行反馈和自动修复能力

EntangleCodec: A Unified Discrete Audio Tokenizer via Semantic-Acoustic Entanglement (Hui Li, Yangfan Gao, Junlin Shang, Changhao Jiang, Tao Gui, Qi Zhang, et al.) 2606.02739 PDF

提升代码生成、执行反馈和自动修复能力。摘要显示:Audio tokenizers serve as the discrete interface between continuous audio and Audio Language Models (ALMs), but existing tokenizers often struggle to support both understanding and generation. 重点核验:任务设置是否真实,是否有代码或数据,评测是否覆盖复杂场景,结论是否能迁移到实际系统。

其他值得关注

阅读边界