一位开发者在Hacker News上发表文章，对当前一代AI产品的设计提出了根本性批评1。他指出，ChatGPT、Claude、Gemini等工具虽然在免责声明中承认自身无法可靠提供信息，但这些警告以灰色小字呈现，实际上只是推卸责任1。作者认为，若不实现一系列关键功能改进，这些AI产品本质上是在欺骗用户，而不是真正提高生产力1。
在具体的功能建议中，作者主张将错误检查作为一等公民特性，通过双列工作表设计强制用户验证每项声明1。他还要求研究工具应将结果呈现为完整引用列表，包含发布日期、作者名称和原文引用，而不是隐藏在小字体链接中1。此外，AI应禁止使用第一人称语言和不必要的道歉1。
代码生成工具的安全问题也是焦点所在1。作者指出这类工具长年来不断破坏用户数据1，呼吁内置沙箱、回滚快照和禁用自动模式等保护机制1。在组织层面，他建议采纳轮班制以防止警觉衰退、进行技能练习以防止技能丧失，以及提供心理健康资源应对AI相关的心理风险1。作者还表示，他从未听到任何人说其AI工作按照严格的生产力指标衡量后效果良好1，暗示现有工具整体上可能无法带来实际收益。
A Hacker News contributor has published a sweeping critique of current AI products, arguing that popular tools like ChatGPT, Claude, and Gemini suffer from fundamental design flaws that fail to address their own inherent limitations 1. The author contends that while these systems include disclaimers acknowledging their unreliability, the warnings are presented in small gray text as legal cover rather than genuine safeguards 1. Rather than improving productivity, the author suggests these products effectively deceive users by glossing over their core shortcomings 1.
The proposed solutions span multiple dimensions of product design. For research and information tools, the author recommends embedding error-checking as a first-class feature through a two-column worksheet approach that forces users to verify each claim against human-reviewed notes 1. Results should be presented as full citation lists with publication dates, author names, and complete text excerpts—not minimal links—to ensure accountability and traceability 1. Additionally, AI systems should abandon first-person language and unnecessary apologies, which waste user time 1. For code-generation tools, the author identifies severe security vulnerabilities, noting that "coding tools keep destroying everyone's data, over the course of years," and calls for mandatory sandboxing, rollback snapshots, and disabled automatic execution modes 1.
Beyond individual features, the author advocates for organizational process reforms: shift-based work schedules to prevent "vigilance decay," skill practice to prevent skill loss, and mental health resources to address psychological risks associated with AI work 1. The underlying concern is whether existing AI tools deliver measurable productivity gains at all, with the author stating: "I have yet to hear from a single person who has said 'yeah, we measured according to your methodology, and it turns out that our AI work is going great'" 1.
评论
还没有评论，欢迎留下第一条。