English

2023 matters because large models begin to shift from spectacular demonstration into infrastructural position.

In the previous year, many people encountered generative AI as surprise: text generation, image synthesis, conversational fluency, and sudden evidence that machines could imitate expressive tasks once considered distinctly human. In 2023, the question changes. The issue is no longer only whether these systems are impressive. The issue is how they can be built into products, workflows, services, and institutions as reusable capability.

This matters because a model stops being merely a destination and starts becoming a layer. It can be called through interfaces, embedded inside software, connected to enterprise systems, fine-tuned for tasks, and incorporated into everyday tools. The large model begins to resemble earlier infrastructure shifts: not just a product people visit, but a capability others build upon.

The significance of this year lies in generalization. A single model can support writing, summarization, search assistance, coding help, customer support, education, translation, and structured reasoning. This does not mean it performs all tasks equally well, but it does mean that one underlying capability begins to spread across many industries.

In China, as elsewhere, this matters because infrastructure quickly reorganizes strategy. Once large models are treated as a general capability layer, companies, developers, platforms, and institutions must decide whether to build, rent, adapt, regulate, localize, or integrate them. The competition moves from isolated novelty toward system placement.

Historically, 2023 is important because it marks the beginning of the model stack era. The model is no longer just an event. It becomes part of architecture.

One-sentence summary:

The key to 2023 is that large models begin to function not only as impressive demos, but as a new infrastructural layer that other systems can build on.

---

中文

2023 年的重要性，在于大模型开始从一种令人震动的演示对象，转向一种基础设施位置。

前一年，很多人第一次接触生成式 AI，感受到的是惊讶：文字能写出来了，图像能生成了，对话能流畅了，机器似乎突然能模仿很多原本被认为很“人类”的表达性任务。到了 2023 年，问题开始变化：人们不再只问“这东西厉不厉害”，而开始问“这东西怎么进入产品、工作流、服务和制度，成为可重复调用的能力层？”

这件事重要，因为模型不再只是一个终点，而开始成为一层。它可以通过接口被调用，被嵌入软件，被接进企业系统，被针对任务微调，被纳入日常工具。大模型开始像此前那些基础设施变化一样，不只是一个被访问的产品，而是一种可被他人继续搭建的能力。

2023 年的意义，在于“通用化”。一个模型可以同时支持写作、总结、搜索辅助、编程帮助、客服、教育、翻译和结构化推理。这并不意味着它在所有任务上都同样优秀，但它意味着：同一种底层能力，开始跨行业扩散，进入许多不同的应用表面。

在中国，这一点尤其重要，因为基础设施一旦形成，就会迅速重组企业和行业的战略。公司、开发者、平台和机构很快都要面对同一个问题：是自建、租用、适配、监管、本地化，还是深度集成？竞争开始从单点新奇，转向系统位置。

从历史上看，2023 年的重要性，在于它标记了“模型栈时代”的开始。模型不再只是一个事件，它开始成为架构的一部分。

如果说 2022 年写的是 AI 作为公共体验突然显形，那么 2023 年写的就是：这种体验如何被迅速抽象为基础设施。模型不再只是给人看一眼“哇”的对象，而开始成为被企业、产品和制度认真接进去的一层能力。

一句话概括：

2023 年的关键，是大模型开始不只作为惊艳演示存在，而开始成为其他系统可以继续搭建的一层新基础设施。