清华大学刘知远团队在 ICLR 2026 上发表的论文《InfLLM-V2: Dense-Sparse Switchable Attention for Seamless Short-to-Long Adaptation》引发了学术界和工业界的广泛关注。这项研究针对 大语言模型 在长上下文处理能力上的瓶颈，提出了创新的解决方案，为现有大模型“热升级”提供了新的思路，有望推动 长文本 处理技术的进步 ...

A recent study suggests Gen Z may be the first generation to lag behind Millennials in key cognitive skills, including attention, memory, problem-solving, and IQ. Experts attribute the decline to ...

At a meeting on the results of the work of BelNPP last November, the ruler proudly noted that one of the results of its construction was "giving impetus to the growth of energy consumption. In 2024, ...

在大语言模型快速迈向更强推理能力与更复杂应用场景的过程中，“上下文长度”已经从一个模型配置参数，演变为制约系统能力上限的关键瓶颈。 一方面，长文档理解、跨轮对话记忆、复杂规划与长链式推理等任务，对模型提出了远超传统 4k 或 8k ...