Don't you mean extinct?
218 points
• 5 days ago
• Article
Link
当 Jurassic Park 于 1993 年上映时，它标志着视觉特效领域的一次重大变革，证明计算机生成图像可以打造出照片级逼真的生物。这一时刻对定格动画大师 Phil Tippett 来说冲击极大，他一度觉得自己毕生的技艺瞬间失去价值。如今，随着大型语言模型（LLMs）进入开发领域，许多程序员也在经历类似的焦虑，担心自动化会让自己的辛苦积累变得无用。然而历史告诉我们，最好的出路不是绝望，而是与新工具共同进化。
了解 LLMs 的工作原理是适应的第一步。像 Andrej Karpathy 的深度解析视频系列，或那些讲述如何从零构建语言模型的技术书籍，能很好地拆解这项技术的运作机制。除了理解底层原理，开发者还要学会把这些"代理"融入日常工作流程。手工敲代码不再是唯一的价值来源，但解决问题的能力依然是核心；与编码相关的那种纪律性仍然必要，尽管入门门槛已经发生变化。
要有效使用 LLMs，就需要调整项目质量管理的方式。生成大量代码很容易，但产出混乱且难以维护的代码是常见陷阱。为此，有经验的开发者越来越倾向于把自己的风格偏好和架构要求形式化为个人配置文件，比如自定义的 Claude 或 Gemini 指令。通过不断打磨这些提示词以强制执行清晰的分层、合理的命名规范等标准，就能确保代理生成的代码保持高质量且便于维护。
AI 辅助开发的兴起也从根本上提高了代码审查的门槛。既然基本语法问题可以由工具解决，就几乎没有理由接受糟糕的提交信息、过度复杂的实现或臃肿的 pull request 。开发者可以先用 LLMs 做初审，确保 PR 在进入人工复核前已简洁、文档完善并配有可靠的单元测试。这种效率甚至让更小、更敏捷的团队也能交付过去可能需要更大团队才能完成的专业级软件。
Phil Tippett 的经历给我们一个关于韧性的有力启示：当 CGI 到来时，他没有退缩，而是拥抱新技术，通过共同开发 Dinosaur Input Device，在传统动画原理与数字工具之间架起了桥梁。他意识到工具会改变，但他对动作、重量感和时机的理解依然不可替代。今天的程序员也可以通过学习新技术并将已有经验应用其中，走出被淘汰的恐惧，继续创造有意义的工作。
When Jurassic Park premiered in 1993, it signaled a seismic shift in visual effects, proving that computer-generated imagery could create photorealistic creatures. This moment was devastating for stop-motion legend Phil Tippett, who felt his life's work had suddenly become obsolete. Many programmers today are experiencing a similar wave of anxiety as Large Language Models enter the development landscape, fearing that the rise of automation will render their hard-earned skills useless. However, history suggests that the best way to move forward is not to despair, but to evolve alongside these new tools.
Learning how LLMs function is the first step toward adaptation. Resources like Andrej Karpathy's deep-dive video series and technical books on building language models from scratch are invaluable for demystifying the technology. Beyond understanding the mechanics, developers must learn how to integrate these agents into their daily workflow. While the act of typing code manually is no longer the sole source of value, problem solving remains the core skill. The discipline traditionally associated with coding is still necessary, even if the barrier to entry has shifted.
Using LLMs effectively requires a shift in how one manages project quality. It is easy to generate vast amounts of code, but producing a messy, unmaintainable output is a common pitfall. To combat this, experienced developers are increasingly formalizing their stylistic preferences and architectural requirements into personal configuration files, such as custom Claude or Gemini directives. By iterating on these prompts to enforce standards like clear layering and proper naming conventions, developers can ensure that the code produced by agents remains high-quality and manageable.
The rise of AI-assisted development has also fundamentally elevated the standards for code reviews. With the burden of writing basic syntax lifted, there is little excuse for poor commit messages, excessive code complexity, or bloated pull requests. Developers can now use LLMs to conduct preliminary reviews, ensuring that PRs are concise, well-documented, and backed by robust unit tests before they reach human eyes. This efficiency even allows for smaller, more agile teams to produce professional-grade software that might have required larger groups in the past.
Ultimately, the story of Phil Tippett offers a powerful lesson in resilience. He did not retire when CGI arrived, but instead embraced the technology, acting as a bridge between traditional animation principles and digital tools by co-developing the Dinosaur Input Device. He realized that while the tools changed, his expertise in movement, weight, and timing remained essential. By learning the new technology and finding ways to apply one's existing experience to it, today's programmers can move past the fear of obsolescence and continue to create meaningful work.
135 comments • Comments Link
- 电影制作的类比突显出一种趋势：在为降低成本和削弱工会势力的推动下，向数字特效的转变最初贬低了实体特效的技艺，但最终却引发了对这些实务技法的怀旧与艺术上的反弹。
- 行业内宣称"实体特效"在现代大片中占主导的说法往往是市场营销手段，许多作品在很大程度上仍依赖隐形的电脑特效（CGI）来补充或替代物理元素。
- 关于拒绝使用大型语言模型会导致程序员落后的说法存在争议；有人认为软件开发的核心在于解决业务问题和设计，而非单纯的产出量，并指出过度追求速度往往会带来长期的维护负担和技术债。
- 资深开发者指出，职业环境中的主要瓶颈很少是打字速度或代码量，而更多是官僚流程、会议、需求不清晰，以及构建现有系统心理模型所需的时间。
- 大型语言模型作为个体开发者的强大倍增器，使他们能够创建以往因耗时而无法实现的复杂工具、测试夹具和可视化工具，从而提升整体流程质量。
- 将大型语言模型引入开发流程可能会加剧团队内部已有的不平衡：如果不通过严格的设计与测试来把控，过分追求速度而忽视质量会导致技术债累积，并产生脆弱、未经充分检验的系统。
- 把程序员置于"适应否则灭亡"的论调，反映了将开发视为工业化产出竞赛的转变；批评者认为这是一种不必要的恐吓策略，并会导致职业技能的退化。
- 人们担忧生成式 AI 工具在大型营利实体中的集中化，这与崇尚开源、独立工具及普遍获取技术的"黑客"价值观形成鲜明对比。
- 对生成式 AI 的实际社会价值存在高度怀疑：有人认为它消耗了宝贵资源、制造了大量低质量的数字副产品，却未能解决迫切的现实文明问题。
- 关于大型语言模型是否最终会超越人类编程能力的争论仍在继续；一种观点认为若 AI 能可靠地产生高效且无 Bug 的代码，人类对可读性和优雅性的要求或将变得多余，另一种观点则认为统计模型不可避免地会产生平庸的"平均"产出。
这场讨论反映了两派之间的深刻分歧：一方把大型语言模型视为职业工程领域至关重要的革命性进化，另一方则将其视为优先产出商品化且平庸结果、损害工艺与质量的破坏性力量。支持者强调在测试、调试和实验性工具方面带来的显著生产力提升，批评者则认为这些工具往往掩盖了行业的结构性问题，例如对良好软件设计的关注不足以及对人类专业知识的贬低。归根结底，这场讨论涉及未来工作形态、 AI 训练的伦理，以及当前软件开发轨迹究竟是在推动真正的进步，还是仅仅制造大量低价值的数字噪音的更广泛焦虑。 • The analogy of filmmaking highlights a trend where a shift toward digital effects, driven by cost-cutting and the de-unionization of labor, initially devalued practical skills, only to eventually trigger a nostalgic and artistic pushback toward practical techniques.
• Industry claims about the dominance of "practical effects" in modern blockbuster films are often marketing tactics, as many productions rely heavily on invisible CGI to supplement or replace physical elements.
• The claim that refusing to use LLMs will cause programmers to fall behind is contested; some argue that software development is about solving business problems and design rather than raw output volume, noting that high velocity often leads to maintenance debt.
• Experienced developers note that the primary bottlenecks in professional environments are rarely typing speed or code volume, but rather bureaucracy, meetings, ambiguous requirements, and the time required to build a mental map of existing systems.
• LLMs serve as powerful force multipliers for individual developers, enabling them to create sophisticated tooling, test fixtures, and visualizers that were previously too time-consuming to justify, thereby improving overall process quality.
• The integration of LLMs can exacerbate existing imbalances in development teams, where a focus on speed over quality might lead to an accumulation of technical debt and brittle, untested systems if not balanced with rigorous design and testing.
• The assertion that programmers must "adapt or die" to LLMs reflects a shift toward viewing development as an industrial output game, a perspective that others criticize as a form of unnecessary fear-mongering and professional de-skilling.
• Concerns exist regarding the centralization of AI tools within large, for-profit entities, which contrasts with traditional "hacker" values that favor open-source, independent tools and universal access to technology.
• There is significant skepticism regarding the actual societal value of GenAI, with some arguing that it produces an abundance of low-quality digital byproducts while consuming precious resources and failing to solve pressing real-world civilizational challenges.
• A debate remains over whether LLMs will eventually surpass human coding ability; one perspective suggests that if AI reliably produces efficient, bug-free code, the human requirements for readability and elegance may become obsolete, while others argue that statistical models will inevitably produce mediocre "average" output.
The discussion reflects a deep schism between those who view LLMs as an essential, revolutionary evolution in professional engineering and those who see them as a destructive force that prioritizes commodified, mediocre output over quality and craft. While proponents emphasize the dramatic productivity gains in testing, debugging, and experimental tooling, critics argue that these tools often mask structural problems in industry, such as a lack of focus on sound software design and the devaluation of human expertise. Ultimately, the conversation touches on broader existential anxieties about the future of work, the ethics of AI training, and whether the current trajectory of software development leads to genuine progress or merely an overwhelming abundance of low-value digital noise.