Fatbobman's Swift Weekly #075
OpenAI Appeals to the US Government: Competitive Concerns Over DeepSeek
OpenAI Appeals to the US Government: Competitive Concerns Over DeepSeek
A few days ago, OpenAI submitted a fifteen-page memorandum to the U.S. government, elevating the competitive threat posed by DeepSeek to a national security concern and attempting to frame it as ideological competition. This submission was OpenAI’s feedback to the White House Office of Science and Technology Policy (OSTP) regarding the publicly solicited Artificial Intelligence Action Plan.
I have never denied the importance of intellectual property rights, nor do I overlook the potential national security risks associated with national-level AI competition. However, for younger companies that have not yet established sufficient credibility regarding intellectual property, actively seeking policy protections might inadvertently weaken their innovative spirit and constrain their long-term growth.
Over the past two years, OpenAI and similar companies have consistently promoted the narrative that building leading large-scale AI models requires enormous financial resources and supercomputing power, setting extremely high barriers that ordinary players cannot overcome. Leveraging this narrative and their early-mover advantage, OpenAI has achieved remarkable valuation and continuously attracted significant investment. However, the emergence of DeepSeek (not the first, nor will it be the last) has disrupted this entrenched perception, prompting more countries, enterprises, and individuals to reconsider the feasibility of entering the large-scale AI arena.
The achievements in today's AI industry, especially in large language models, are the result of collective efforts and the open sharing of numerous researchers and enterprises. As a company benefiting from the open-source community and actively sharing its own research findings and technical details, DeepSeek has given back to the community and contributed to lowering AI training and inference costs. Such open outcomes should benefit the entire society, including existing leading AI companies. However, as I pointed out in Issue 68 of my newsletter: "For major AI companies accustomed to high investments and large-scale resource allocation, it will undoubtedly be challenging to shift their mindset in the short term. Even if DeepSeek’s methods provide some inspiration, without fundamental changes in philosophy, these companies will struggle to achieve sustained and significant progress in reducing training costs."
Perhaps OpenAI is currently trapped in this mindset transition dilemma. Realizing the difficulty of rapidly improving its cost structure, it has chosen more aggressive competitive methods, seeking external intervention to slow down its competitors. While this strategy might temporarily alleviate pressure from capital markets, it ultimately does not resolve the company's underlying developmental bottlenecks.
The rise of AI is not only a significant technological shift in human history but also a commercial opportunity filled with challenges. As a promising industry with vast market potential, fierce competition is inevitable as countries and enterprises accelerate strategic positioning. As large-model technologies approach performance ceilings and differences in model capabilities narrow, offering distinctive application scenarios and cost-effective products will be critical factors determining market success. OpenAI may not have adopted an optimal competitive strategy—unless it candidly confronts and effectively addresses its structural issues, its true replacements in the future may not be DeepSeek, as claimed, but rather those domestic competitors currently behind but stronger in terms of cost-efficiency and unique application scenarios.
DeepSeek is just one among many new challengers reshaping the industry landscape, serving as evidence of the increasing feasibility for more participants to enter this "high-barrier" industry. Soon, we will likely witness similar DeepSeek-style breakthroughs emerging globally.
Since its inception, ChatGPT has attracted me through its inherent strengths and outstanding performance. I sincerely hope it continues to retain its users through continuous innovation and excellence, rather than relying on external measures like policy protections.
Previous Issue｜Newsletter Archive
If you appreciate my work and want to promote your product to the Swift and iOS developer community, sponsoring my blog & newsletter could be an excellent opportunity for you.
Original
Key Considerations Before Using SwiftData
At the recently concluded Let’s Vision 2025 conference, I received numerous questions about SwiftData: "Is SwiftData mature enough for real-world projects?" and "As a beginner, how can I effectively use SwiftData?" These questions reflect developers' strong interest in Apple's latest data persistence framework, but they also reveal some hesitation when making technical decisions. This article aims to serve as a guide for developers interested in SwiftData, helping you understand its advantages and limitations so you can make informed choices based on your project needs.
Recent Recommendations
Behind the Scenes of Async Functions
Swift Concurrency introduces a safer and more efficient asynchronous programming model, but do you truly understand how it works under the hood? In this article, Vitaly Batrakov provides an in-depth exploration of core concepts such as async/await, Tasks, Jobs, Executors, Actors, and the Cooperative Thread Pool. This article is ideal for developers with some concurrency experience, helping you build a clearer mental model of how Swift Concurrency operates.
When the Swift Compiler Deleted Code in Stdlib
Swift’s compiler optimizations are designed to improve performance, but sometimes they can go too far and introduce unexpected bugs. In this article, WeZZard shares a use-after-free crash case that occurred under the -Osize optimization level, where a critical piece of code was erroneously eliminated by the Swift compiler, leading to memory access issues. Fortunately, after thorough investigation, WeZZard submitted a Pull Request that has been successfully merged, fixing the issue. While this article delves into some advanced topics, if you're interested in Swift compiler optimizations, SIL, and LLVM, it's definitely worth a read!
Swift Testing Completion Handlers
In XCTest, we typically use XCTestExpectation to wait for asynchronous operations to complete. However, Swift Testing does not provide a direct equivalent to expectation. Instead, the recommended approach is to use continuations to convert completion handler-based asynchronous code into async/await. While this transition is straightforward for new test cases, migrating a large number of existing XCTest cases can be quite time-consuming. In this article, Keith Harrison shares an efficient migration strategy using withCheckedContinuation, streamlining the process and reducing redundant work to significantly improve productivity.
SwiftUI: Interactive Charts
Unlike many other charting frameworks, Swift Charts employs a declarative API, which may feel unintuitive when implementing interactive features. In this article, Itsuki provides a deep dive into the built-in interaction mechanisms of Swift Charts and explores how to leverage chartGesture(_:) for custom gestures, enabling more flexible and intuitive user interactions.
Under the Hood: SwiftUI
SwiftUI makes UI development more intuitive, but its black-box nature makes it difficult for developers to understand how it operates under the hood. In this article, Mihai Popa adopts an interview-style narrative to clearly and concisely explain key concepts such as SwiftUI’s declarative rendering architecture, state-driven updates, view diffing mechanisms, and its bridging with UIKit.
Browse No More
Once upon a time, browsing the internet was an adventure—clicking links, stumbling upon hidden gems, and discovering unique voices along the way. However, AI-powered answer engines such as ChatGPT, Perplexity, Grok, and Gemini are reshaping how we access information. While they provide instant and precise responses, they are also quietly diminishing our agency and reducing the diversity of the web. Instead of exploring freely, we are now guided by AI-selected answers, missing out on independent creators, niche communities, and fresh perspectives. In this article, Paul Stamatiou reflects on how this shift is affecting the internet ecosystem and explores how AI-driven personalization could help restore the joy of discovery.
OpenAI 向美政府状告 DeepSeek：他不讲武德！
几天前，OpenAI 向美国政府提交了一封长达十五页的进言，将 DeepSeek 带来的竞争威胁上升至国家安全层面，并试图将其框定为意识形态竞争。这是 OpenAI 对美国白宫科技政策办公室（Office of Science and Technology, OSTP）就 人工智能行动计划（AI Action Plan）公开征求意见的反馈。坦率而言，当看到这则新闻时，我不禁哑然失笑——难以想象行业巨头 OpenAI 会如此沉不住气，展现出这般脆弱的一面。
这两年来，OpenAI 等企业一直向全球灌输一种观念：打造领先的大模型必须依靠海量资金与超级算力，这是一个门槛极高的领域，普通玩家根本无缘参与。凭借这一论调和先发优势，OpenAI 创造了惊人的估值，并源源不断地吸引资本涌入。然而，DeepSeek 的出世打破了这一既定认知，促使越来越多的国家、企业乃至个人开始重新审视进入大模型领域的可能性。
OpenAI 显然察觉到了投资者对其“高门槛论”的迟疑与动摇。为了稳固自身地位，他们选择采取更为强硬的手段，急于重申资金的决定性作用，并巧妙地为自身对资本的渴求披上了一层“意识形态对抗”的外衣。
AI 行业，尤其是大模型领域今天取得的成就，源自众多科研人员和企业的共同努力与无私分享。作为一家受益于开源社区，并积极公开自身研究成果和技术细节的企业，DeepSeek 推动了社区的发展，也为降低 AI 训练和推理成本作出了贡献。这些开放成果本应惠及整个社会，当然也包括那些已处领先地位的 AI 公司。但正如我在第 68 期周报 中所指出的：“那些习惯于高投入、大规模资源配置的头部 AI 企业，在短期内转变思维模式无疑是困难的。即使 DeepSeek 的方法能够提供一些启发，但如果没有彻底的理念变革，这些企业在降低训练成本上将难以取得持续的显著进展。”
显而易见，OpenAI 正陷入这种思维转型的困境。当意识到自身难以快速改善成本结构时，它选择了更为激进的竞争手段，即向美国政府寻求外部干预，以此减缓竞争对手发展的步伐。这种幼稚的做法像极了小孩子在向家长告状，或许能在短期内缓解资本市场的压力，但从长远来看，并不能解决企业自身的发展瓶颈。
AI 的崛起不仅是人类科技史上的一次重大变革，也是一场充满商业机遇的挑战。作为拥有广阔市场前景的朝阳产业，激烈的竞争在所难免，各国与企业纷纷加紧布局抢占先机。在大模型技术逐渐接近性能天花板，各类模型性能日益趋同的当下，能够提供特色应用场景和具备更高性价比的产品，才是真正决定市场胜负的关键。OpenAI 此番或许并未采用合适的竞争策略——如果不能坦诚面对并有效解决自身存在的结构性问题，那么未来真正能取而代之的，很可能并非它所指控的 DeepSeek，而是在性价比与特色场景方面更具优势的那些曾被其甩在身后的国内竞争者。
值得一提的是，与 OpenAI 相比，Anthropic 向白宫提交的反馈甚至更为极端。DeepSeek 不过是众多挑战现有格局的新锐之一，它只是让更多人看到了进入这个所谓“高门槛”行业的可能性。相信很快我们会看到来自全球各地更多类似 DeepSeek 式的成果展现。
我从不否认知识产权的重要性，也不否认国家级 AI 竞争可能带来的安全风险。然而，对于那些自身尚未在知识产权问题上建立足够可信背书的年轻企业来说，积极寻求政策保护，反而可能削弱自身的创新动力，限制长期发展。
如果您发现这份周报或我的博客对您有所帮助，可以考虑通过 爱发电，Buy Me a Coffee 支持我的创作。
原创
SwiftData 使用前必须了解的关键问题
在刚刚结束的 Let’s Vision 2025 大会上，我收到了许多关于 SwiftData 的提问：“SwiftData 是否已经足够成熟，可以用于实际项目？”、“作为初学者，如何高效地使用 SwiftData？”。这些问题反映了开发者对苹果最新数据持久化框架的浓厚兴趣，但也透露出技术选型时的犹豫。本文旨在为对 SwiftData 感兴趣的开发者提供一份指南，帮助你了解 SwiftData 的优势与局限，并根据项目需求做出明智的技术选择。
近期推荐
Async 函数的幕后机制 (Behind the Scenes of Async Functions)
Swift Concurrency 带来了更安全、更高效的异步编程模型，但你真的理解它的内部运作吗？在这篇文章中，Vitaly Batrakov 深入解析了 async/await、Tasks、Jobs、Executors、Actors 以及 Cooperative Thread Pool 的核心概念，帮助开发者建立更清晰的 mental model。本文适合已有并发基础的开发者，能帮助你更全面地掌握 Swift Concurrency 的工作原理。
当 Swift 编译器在标准库中删除代码时 (When the Swift Compiler Deleted Code in Stdlib)
Swift 编译器的优化机制本意是提升性能，但有时候却可能“优化过头”，引发意想不到的 Bug。WeZZard 在这篇文章中分享了一起 use-after-free 崩溃案例，该问题发生在 -Osize 优化级别下，根本原因是 Swift 编译器误删了关键代码，导致程序访问已释放的内存。幸运的是，WeZZard 经过深入分析后，在 Swift 社区提交的 Pull Request 已成功合并，修复了这个问题。这篇文章或许读起来有些硬核，但如果你对 Swift 编译器优化、SIL 以及 LLVM 感兴趣，它绝对值得一读！
Swift Testing 中的 Completion Handler 处理 (Swift Testing Completion Handlers)
在 XCTest 中，我们通常使用 XCTestExpectation 来等待异步操作完成，但 Swift Testing 并没有提供直接等效的 expectation 机制。官方推荐的做法是使用 continuations 来将基于 completion handler 的异步代码转换为 async/await。如果你是从零编写测试，这种方式非常自然，但当你需要迁移大量 XCTest 代码时，工作量可能会大幅增加。Keith Harrison 在本文中分享了一种高效的迁移策略，利用 withCheckedContinuation 简化 completion handler 代码的转换，让你在迁移过程中减少不必要的重复工作，大幅提升效率。
🪜 SwiftUI：交互式图表 (SwiftUI: Interactive Charts)
与许多常见的图表框架不同，Swift Charts 采用了声明式 API，这让某些开发者在实现交互时感到不太直观。在本文中，Itsuki 深入解析了 Swift Charts 的内置交互方式，并探讨了如何使用 chartGesture(_:) 来自定义手势，从而实现更灵活的用户交互体验。
🪜 SwiftUI 的底层原理 (Under the Hood: SwiftUI)
SwiftUI 让用户界面开发变得更加直观，但由于其黑盒特性，开发者往往难以洞悉其底层运作机制。在这篇文章中，Mihai Popa 通过一种访谈式的叙述方式，清晰而简洁地阐释了 SwiftUI 的诸多核心原理，包括声明式渲染架构、状态驱动的界面更新机制、视图差异化比较（diffing）算法，以及与 UIKit 的桥接技术等关键内容。
互联网的探索感正在消失 (Browse No More)
曾几何时，我们在互联网上的浏览充满了探索的乐趣，随意点击、偶然发现，让每次搜索都可能成为一场意想不到的冒险。然而，如今 AI 答案引擎（如 ChatGPT、Perplexity、Grok、Gemini 等）正在重塑我们的信息获取方式——它们提供即时、精准的回答，却在无形中削弱了我们的自主性，让互联网的多样性逐渐消失。我们不再主动探索，而是被 AI 精选的答案所引导，错过了许多原本可能发现的独立创作者、小众社区和新鲜视角。在这篇文章中，Paul Stamatiou 深入探讨了这种变化对互联网生态的影响，并提出 AI 未来可能的个性化发展方向，以帮助我们找回探索的自由。