arXiv · cs.AI · Xujia Wang, Yixuan Liu, Shaohan Huang, Furu Wei, Lei Hou, Bin Xu, Juanzi Li · 4 小时前模型评分42
WebCatalyst：用高保真强化学习提升交互式 HTML 生成能力
Catalyzing Interactive HTML Generation Capabilities via High-Fidelity Reinforcement Learning
导读
研究团队提出浏览器交互强化学习框架 WebCatalyst，用游戏级 GUI 评估器、Visual Anchor 和 Action-Trace Credit 三个组件提升交互式 HTML 生成质量。
分类置信度
中 · 待人工复核
- 只有单一媒体来源，需要人工复核。
来源级别：媒体与个人
事件类型：研究发布
证据摘录
- “We introduce WebCatalyst, a browser-interaction reinforcement learning framework with three complementary components”
- “Trained on fewer than 1K requests, predominantly for games, WebCatalyst raises the Cookie-Bench Mini Games score from 68.8 to 75.3 and also the WebGen-Bench accuracy from 48.8 to 56.9 on Qwen3.6-35B-A3B, outperforming Claude Opus 4.7 in average rank across evaluated metrics.”
来源：arXiv · cs.AI · arxiv.org