Primary sourcearXiv cs.AI
View sourceEvent ·Multi-source
Gym-Anything-RT 发布:评估视觉语言智能体的实时交互能力
arXiv 论文介绍 Gym-Anything-RT,一个用智能体流水线自动生成视觉小游戏的环境,用于测试智能体在环境状态持续变化下的实时交互。它基于六项实时能力分类构建了 100 个实时小游戏,并支持暂停或减速环境做对照。结果显示实时场景下能力差距明显:GPT-6 Astra (xHigh) 实时模式通过率仅 9.0%,暂停环境下升至 75.0%。论文还观察到前沿智能体在被允许写代码时会绕过实时感知与操作,环境与小游戏已公开。
Current read
研究者做了100个实时小游戏测智能体,GPT-6 Astra实时模式才9%通过率,一暂停就飙到75%,差距很直观。
10 sources64° AI heatLast updated 2026/10/27 17:13:05
Evidence
10 sourcesRelated source 1AIGCLINK
View sourceRelated source 2Simon Willison’s Weblog
View sourceRelated source 3marktechpost
View sourceRelated source 4rohanpaul_ai
View sourceRelated source 5OpenAI Blog
View sourceRelated source 6IT之家
View sourceRelated source 7Shubham Saboo
View sourceRelated source 8Decoder
View sourceRelated source 9Artificial Analysis
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.