ArticleHongbo Jin, Mingnan Zhu, Jingqi Tian, Xu Jiang, Zhongjing Du, Haoran Tang, Siyi Xie, Qiaoman Zhang, Jiayu DingResearch
Practical AI workflows such as in-context retrieval and RAG depend on a model's ability to reliably process task-specific data supplied at runtime. The low success rate of current frontier models on context benchmarks highlights a major bottleneck for enterprise adoption.
The authors identify a clear divergence between static knowledge reasoning and dynamic context comprehension. The CL-Bench metric demonstrates that simply increasing model size or prompt length does not automatically solve context-dependent reasoning, motivating targeted synthesis methods like Context-CoT.
Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.
Similar in content, found by the AI search.
Article:Jiarui Lu, Thomas Holleis, Yizhe Zhang, Bernhard Aumayer, Feng Nan, Felix Bai, Shuang Ma, Shen Ma, Mengyu Li, Guoli Yin, Zirui Wang, Ruoming Pang
ToolSandbox is an evaluation benchmark that assesses LLM tool-use capabilities using stateful environments and multi-turn interactive dialogues.
Article:Lei Liang et al.
A research team led by Lei Liang introduces Knowledge Augmented Generation (KAG), a framework extending conventional RAG with knowledge graphs to address logic, temporal, and numerical reasoning deficits in specialised domains.
Article:Anthropic
Anthropic details the shift from prompt engineering to holistic context engineering for autonomous AI agents. Because language models operate within finite attention budgets dictated by transformer architectures, developers must carefully curate tokens and adopt just-in-time retrieval strategies.
Article:Siru Ouyang, Jun Yan, I-Hung Hsu, et al.
ReasoningBank is a memory framework for LLM agents that distils generalisable reasoning strategies from self-judged successful and failed experiences, enabling continuous self-evolution over time.
Link:Subquadratic Inc.
AI research company Subquadratic has introduced SubQ, a model designed for multi-million token reasoning. By using a proprietary algorithm called Subquadratic Sparse Attention (SSA), the company aims to significantly cut compute and memory demands compared to traditional Transformer-based architectures on large contexts.
Article:Peiyan Li, Yuze Zhu, Yixiang Chen, Qisen Ma, Yuan Xu, Jiabing Yang, He Guan, Yan Huang, Hongtao Wu, Xiao Ma, Tao Kong, Liang Wang, Tieniu Tan
Researchers introduce BridgeVLA++, an upgraded vision-language-action framework designed for 3D robotic manipulation. It introduces a unified spatio-temporal memory architecture into pre-trained vision-language models, enabling persistent spatial reasoning and temporal history tracking without losing data efficiency.
Parts of this content were created with the assistance of AI.
