📄 中文摘要
Ilya Sutskever 表示扩展时代已经结束，而 Yann LeCun 则以 10 亿美元押注大语言模型（LLMs）是死胡同。在这种背景下，五种新兴的范式正在逐渐取代单纯的规模扩展。这些新架构不仅在技术上具有重要意义，还可能对开发者的技术栈产生深远的影响。每种架构都有其独特的特征和应用场景，了解这些变化对于保持技术前沿至关重要。文章提供了对每种架构的开发者友好的解析，包括其定义、重要性以及深入学习的资源链接。
五种架构取代粗暴的 AI 扩展（及其对你的技术栈的影响）
📄 English Summary
5 architectures replacing brute-force AI scaling (and what they mean for your stack)
Ilya Sutskever has declared the era of scaling over, while Yann LeCun has bet $1 billion that large language models (LLMs) are a dead end. In this context, five emerging paradigms are converging to replace the brute-force approach of simply making models larger. These new architectures not only hold significant technical implications but also have the potential to profoundly impact developers' tech stacks. Each architecture has its unique characteristics and application scenarios, making it crucial to understand these shifts to stay at the forefront of technology. The article provides a developer-friendly breakdown of each architecture, including its definition, importance, and links to deeper learning resources.