Expired50% confidenceOpinionExact time
Mainstream large language models and code generation tools have a distributional bias in UI code training data, tending to reproduce patterns from high-weight sources such as Tailwind CSS documentation, shadcn/ui, and Material UI.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/13/2026
First Seen
Valid until: 10/11/2026(expired)
Sources
BaseUI：88个组件的「反同质化」React开源库，对抗AI界面千篇一律
redditr/opensource7/12/2026
Related Claims
UnverifiedHTML是AI训练数据中占比最大的格式之一，模型对HTML语法和语义的掌握程度远超任何专有格式65% similarUnverifiedTraditional browser automation frameworks are designed for software testing and data scraping; faced with low signal-to-noise DOM tree structures, agents must consume significant inference resources to locate page elements, and frontend redesigns can easily cause scripts to break.64% similarUnverifiedLLMs have far more training data on HTML/CSS than on React DSL, so the quality and stability of generating video content using the HTML approach is higher.63% similarUnverified对于弹层、提示气泡、下拉菜单等常见UI组件，纯CSS方案在代码量、可访问性和维护成本上比JavaScript方案具备优势62% similarUnverified基于语义属性（aria-label、role、text content）生成的选择器比 XPath 或 CSS 路径对 UI 变更更具鲁棒性61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/502711API
curl https://kongchang.com/api/v1/knowledge/claims/502711MCP
get_claim(id=502711)