📄 中文摘要
软件工程的历史大部分时间里，系统构建基于一个简单且令人安心的假设：相同的输入会产生相同的输出。然而，随着人工智能技术的发展，这一假设面临挑战。AI 系统的输出往往是非确定性的，受多种因素影响，包括数据的变化和模型的复杂性。这种非确定性使得传统的测试策略和工具难以适用，开发者需要重新思考如何设计和验证这些系统。文章探讨了如何应对这些挑战，强调了理解和管理非确定性依赖的重要性，以确保 AI 系统的可靠性和有效性。
AI 不是一个库：为非确定性依赖设计
出处: AI Is Not a Library: Designing for Nondeterministic Dependencies
发布: 2026年2月18日
📄 English Summary
AI Is Not a Library: Designing for Nondeterministic Dependencies
For much of software engineering history, systems have been built on the comforting assumption that the same input will yield the same output. However, with the rise of artificial intelligence, this assumption is being challenged. AI systems often produce nondeterministic outputs influenced by various factors, including data variability and model complexity. This nondeterminism complicates traditional testing strategies and tools, prompting developers to rethink how to design and validate these systems. The article emphasizes the importance of understanding and managing nondeterministic dependencies to ensure the reliability and effectiveness of AI systems.