Verified65% confidenceFactExact time
Google DeepMind在开发Gemini时采用了从头开始的多模态训练策略，使其能够直接理解视频中的时序变化和音频中的音调情感等非文本信息
3
Sources
65%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026
Sources
How to Choose Among 10 AI Tools? A Practical Guide Organized by Use Case
bilibili中山龙阿介6/21/2026
Related Claims
UnverifiedGoogle DeepMind在多模态模型的GUI理解和交互方面持续投入，包括利用大规模网页截图数据训练模型以及探索Android设备操控能力80% similarUnverifiedGoogle DeepMind提出了Nested Learning（嵌套学习）相关方法80% similarUnverifiedGoogle DeepMind的Genie项目尝试从无标注视频中学习可交互的世界模型78% similarUnverifiedDeepMind在VLA上叠加了一层Gemini模型，用户可直接用自然语言下达指令，机器人能一边工作一边对话74% similarUnverifiedGoogle DeepMind的SIMA项目训练了一个能够在多款3D游戏中执行自然语言指令的通用智能体74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/41265API
curl https://kongchang.com/api/v1/knowledge/claims/41265MCP
get_claim(id=41265)