arXiv:2608.29387 · 评测基准
EvoGenUI-Bench: Evaluating LLMs as Multi-Turn Generative UI Assistants
EvoGenUI-Bench：评估 LLM 作为多轮生成式 UI 助手
EvoGenUI-Bench: Evaluating LLMs as Multi-Turn Generative UI Assistants
- 类型：arxiv
- 标识：2608.29387
- 链接：https://arxiv.org/abs/2608.29387
- 主分类：evaluation
- 形态：benchmark
- TLDR：Large language models can generate interactive web interfaces, but reliable generative UI requires maintaining an executable artifact as user requests evolve. We introduce EvoGenUI-Bench, a benchmark for multi-turn interface maintenance comprising 150 five-turn tasks and 750 turns across three scenarios: information presentation, executable interaction, and tool-grounded external state. We execute generated artifacts in a browser and evaluate them using screenshots, source and DOM evidence, actor traces, and runtime logs. Beyond turn-level and episode-level success, we measure cross-turn reten
- 待LLM分类：否
- 标题中文：EvoGenUI-Bench：评估 LLM 作为多轮生成式 UI 助手
- TLDR中文：大语言模型能够生成可交互的 Web 界面，但可靠地生成式 UI 需要在用户请求不断变化时维护一个可执行产物。我们提出 EvoGenUI-Bench，一个面向多轮界面维护的基准，包含 150 个五轮任务、覆盖三个场景共 750 轮：信息呈现、可执行交互以及基于工具的外部状态。我们在浏览器中执行生成的产物，结合截图、源码与 DOM 证据、执行轨迹与运行时日志进行评估。除逐轮与整轮次的成功率外，我们还衡量跨轮内容留存
- 来源文件：
- /inbox/tom/_candidates/2026-09-02-agent-rag-longcontext-candidates.json