Google Researchers Propose StreamVibe for Proactive Video Agents to Understand Evolving User Intent
Google researchers and collaborators introduced StreamVibe, a framework for proactive video agents that observe continuous video streams to identify explicit, implicit, and dynamically evolving user needs in real-world interactions. The team also built an evaluation and training system covering these intent types, aiming to move agents from passive instruction-following to proactive assistance that adapts as tasks progress.
Coverage timeline
机器之心机器之心
近期，Meta 的 Muse、OpenAI 的 Dot 等产品相继推出，让 Personal Agent（个人智能体）再次成为大模型领域的关注焦点。相比传统的对话助手，这类 Agent 不再局限于回答问题，而是能够结合用户的个人背景、历史记忆与应用上下文，主动执行任务并持续推进长期目标。一个更具主动性、能够长期陪伴用户的 AI 助手，正在逐渐成为现实。 然而， 长期记忆只是 Personal Agent 的起点。记住用户过去做过什么，并不代表能够理解用户此刻需要什么 。在真实场景中，用户往往不会明确说出每一个需求。做菜时反复查看步骤、操作软件时遇到困难，或任务进行中临时改变计划，这些细微的行为与环境变化，都可能暗示着新的协助需求。如何从持续变化的环境中捕捉这些信号， 理解用户尚未表达的意图，并决定何时、如何提供帮助 ，是 Personal Agent 走向真实世界需要解决的关键问题。 围绕这一问题， Google 等研究团队提出 StreamVibe ，聚焦动态环境下的主动式视频智能体。研究通过持续的视频观察与交互，让 Agent 不仅能够识别用户的显式需求，还能发现隐藏在行为中的潜在意图，并随着任务进展及时更新对用户目标的理解。团队进一步构建了覆盖 显式、隐式及动态演变需求的评测与训练体系 ，探索如何让 Agent 从被动等待指令，走向能够持续理解用户、适时主动介入并灵活调整协助方式的智能伙伴。 论文标题：StreamVibe: Proactive Video Agents for Evolving User Intent 论文地址：https://arxiv.org/pdf/2610.34662 项目主页：https://stream-vibe.github.io/ 现实交互中的用户意图会持续变化 “水开了提醒我”和“帮我把这道菜做好”，对助手提出的要求并不相同。前者已经给出明确的触发条件，后者只描述目标，需要助手结合正在发生的事情，判断何时介入、提供什么帮助。 真实任务还会不断产生新信息。用户可能已经完成某一步，刚才有用的建议此时就成了重复提醒；也可能临时修改要求，原先的计划随之失效。即使保存了完整历史，助手仍需要分清：哪些要求仍然有效， 哪些步骤已经完成，眼前的新情况意味着什么。 StreamVibe 因而把主动视频交互放进持续变化的用户—助手—环境关系中。
Related stories
Google Cloud unveils universal Gemini agent for multi-day enterprise workflows
At the Gemini at Work 2026 event, Google Cloud announced the Gemini agent, a universal office agent that runs in the cloud and can handle multi-day tasks across Workspace, Microsoft 365, and Slack. The agent can plan and execute tasks, delegate to subagents, use multiple AI models including Anthropic's Claude, and has its own workplace identity with an email address and calendar.
4 reports · 4 sources
VideoGen-Agent: 8B multimodal agent learns tool use for video generation via RL
Researchers released VideoGen-Agent, an 8B multimodal agent built on Qwen3-VL-8B-Instruct that learns to call retrieval, generation, and verification tools for video generation through supervised fine-tuning and multi-task reinforcement learning. On the VABench benchmark with 600 prompts covering six capabilities, its Toolset 1 configuration scored 75.6, up 19.1 from the base generator's 56.5, and reached 86.1 when generation tools were upgraded without changing agent parameters. In human comparisons, evaluators preferred the upgraded configuration's videos over the strongest single-model baseline in 84.3% of judgments.
1 reports · 1 sources
Google Rolls Out Personal Intelligence for Gemini in Cars
Google is rolling out Personal Intelligence for Gemini on Android Auto now and to cars with Google built-in in the coming months as an opt-in feature. The update lets Gemini access user data across Google apps like Gmail, Calendar, Maps, Photos, YouTube, and Search, enabling tasks such as checking flight status from Gmail confirmations or navigating to appointments by voice.
2 reports · 2 sources
Ant Group's Bailing open-sources Ling-3.1-flash-1019 for personal agents
On October 28, 2026, Ant Group's Bailing open-sourced Ling-3.1-flash-1019, an updated model for personal-agent tasks, along with its base model Ling-3.1-flash-base. The new version scores 44 on the Artificial Analysis Intelligence Index v4.3.2, up 3 points from the previous version, and shows improvements across benchmarks including AA-Briefcase, GDPval-AA, AutomationBench-AA, CritPt, and SciCode. The release is linked to the Lingguang 2.0 upgrade, which positions the assistant as a proactive personal AI that can manage long-running tasks.
News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe