Vitali Friedman 挑战当前的 AI 交互范式，认为用户不应学习提示工程，设计师应创建更直观的界面，如按钮和滑块，而非依赖文本框。
Vitali Friedman challenges the current AI interaction paradigm, arguing that users shouldn't have to learn prompt engineering and that designers should create more intuitive interfaces like buttons and sliders instead of relying on text boxes.
要点 · TL;DR
AI 应主动提问澄清以减少无效交互。 AI should ask clarifying questions to reduce wasted iterations.
用按钮、筛选器等熟悉控件替代单一文本框。 Use familiar UI controls like buttons and filters instead of just a text box.
展示来源和置信度以建立对 AI 输出的信任。 Show sources and confidence levels to build trust in AI outputs.
核心观点 · Key points
AI应在生成输出前先提问澄清，以减少无效迭代。 AI should ask clarifying questions before generating output to reduce wasted iterations.
使用按钮、筛选器和滑块等熟悉控件，而非仅用文本框。 Use familiar UI controls like buttons, filters, and sliders instead of just a text box.
展示来源和置信度以建立对AI输出的信任。 Show sources and confidence levels to build trust in AI outputs.
设计输出后优化功能：允许编辑、排序和筛选结果。 Design for post-output refinement: allow editing, sorting, and filtering of results.
优先考虑人类需求而非AI炒作；专注于降低交互成本。 Prioritize human needs over AI hype; focus on reducing interaction cost.
反共识 · Contrarian takes
提示工程不应成为用户技能；AI应适应用户。 Prompt engineering should not be a user skill; AI should adapt to users.
在用户输入提示前放慢节奏，可带来更好结果并减少浪费。 Slowing down users before they prompt leads to better outcomes and less waste.
AI优先设计有误；应以人类优先，AI作为组件。 AI-first design is misguided; focus on human-first with AI as a component.
可见的AI标签可能降低信任；将AI静默集成到工作流中更好。 Visible AI labels can reduce trust; quiet AI integrated into workflows is better.
智能体被过度炒作；可靠性和护栏仍是主要挑战。 Agents are overhyped; reliability and guardrails remain major challenges.
本期章节 · Chapters（共 21）
引言与魔法盒体验Introduction and the magic box experience
AI反问问题AI asking questions back
引言与推荐观看视频Introduction and recommendation to watch video
深度研究模式与不必要阐述Deep research mode and unnecessary articulation
更好方法：Perplexity的澄清与任务构建模式Better approach: Perplexity's clarification and task builder pattern
共识：过滤器与共识计量器Consensus: filters and consensus meter
AI体验中的过滤器与UIFilters and UI in AI experiences
准确性及Elicit的星号Accuracy and Elicit's asterisks
AI融入产品结构AI integrated into product fabric
放缓提示以提升输出Slowing down prompting for better output
编辑AI输出并减少摩擦Editing AI output and reducing friction
双重检查响应与来源验证Double-check response and source verification
通过准确性与范围构建信任Building trust through accuracy and scoping
AI中的记忆与个性化Memory and personalization in AI
AI输出的交互式优化Interactive refinement in AI outputs
Exa的结构化数据与过滤Exa's structured data and filtering
赞助信息与引言Sponsor message and introduction
安静AI与可见AIQuiet AI vs visible AI
关注用户需求与交互成本Focus on user needs and interaction cost
定价层级限制与用户引导Limitations of pricing tiers and guiding users