🏠較早收集於 83m

Gemini 手機多步驟任務功能上線

Gemini 手機多步驟任務功能上線
PostLinkedIn
🏠閱讀原文: IT之家
#agentic-ai#on-device-detection#android-geminigoogle-gemini-android-featuresgooglegeminigalaxy-s26pixel-10

💡Gemini's phone agents automate rides/shopping—new benchmark for mobile LLM apps

⚡ 30-Second TL;DR

有什麼變化

Gemini 後台自動化多步驟任務(如叫車),用戶可即時監控

為什麼重要

推動手機代理式 AI,實現日常任務自動化並提升端側安全。開發者可整合新 Gemini 工具於 Android。

下一步行動

Prototype Gemini multi-step task delegation in your Android app using the beta rollout on Pixel 10.

誰應關注:Developers & AI Engineers

關鍵要點

  • Gemini 後台自動化多步驟任務(如叫車),用戶可即時監控
  • Circle to Search 升級多物件偵測與服裝虛擬試穿
  • Gemini 端側詐騙偵測,警示通話與訊息詐騙模式
  • 初始美韓食品/雜貨/叫車 App,限 Galaxy S26/Pixel 10
  • 電話詐騙偵測預設關閉,聯絡人通話不啟用

🧠 深度解析

背景與延伸:來自公開資料,非原文內容。引用 8 個來源。

🔑 增強重點摘要

  • Multi-step automations execute in a secure virtual environment (sandbox) on the device, limiting Gemini's access to only specified apps and preventing broader data exposure.
  • Scam Detection for phone calls, powered by Gemini, previously launched on Pixel devices in the US, Australia, Canada, India, Ireland, and UK before expanding to Galaxy S26 in the US.
  • Gemini scam text detection uses on-device models and is initially available on Pixel 10 series in the US, Canada, and UK, with upcoming support for Galaxy S26.
  • Feature rollout is in beta with real-time user monitoring, explicit initiation required, and ability to intervene or stop tasks during execution.
📊 競品分析▸ Show
FeatureGoogle Gemini (Android)Apple Siri (iOS)OpenAI ChatGPT
Multi-step Task AutomationBeta for rideshare/food/grocery on Pixel 10/Galaxy S26 (US/Korea); on-device sandbox [1][2][4]Teased at WWDC 2025, limited beta with third-party app integration [1][3]Scheduled tasks, agent for computer tasks like calendar/slideshow/code; not mobile app-focused [4]
Scam DetectionOn-device for calls (Pixel/S26) and texts (Pixel 10); English/US initial [2][4]Not mentioned in sourcesNot applicable
PricingFree with eligible devicesFree with iOS devicesSubscription-based (ChatGPT Plus)
BenchmarksLower latency via on-device; 3B+ Android devices [1][5]Beta limitations; smaller iOS baseCloud-dependent; broader agent testing [1][3]

🛠️ 技術深入

  • Automations leverage Android's deep OS integration, likely using accessibility APIs for navigating third-party app UIs, handling authentication, and applying user preferences without constant input.
  • On-device execution in a constrained sandbox ensures privacy, low latency, and limited app access, distinguishing from cloud-heavy competitors.
  • Scam detection employs local Gemini models for real-time analysis of call patterns and text content, processing entirely on-device.

🔮 前景展望AI analysis grounded in cited sources

Gemini multi-step tasks will expand beyond initial US/Korea and app categories by end of 2026
Sources indicate beta status with plans to extend scam detection to more devices and regions, signaling phased global rollout for automations.
Android's 3B+ device base positions Google to lead consumer AI agents over iOS
Google's on-device focus and massive install base provide distribution advantage while Apple's features remain in limited beta.
On-device AI will drive higher daily user reliance on Gemini vs cloud assistants
Features emphasize faster responses, privacy, and managed autonomy in real mobile apps, aligning with mobile user expectations.

時間線

2025-06
Apple previews Siri multi-step automation at WWDC, influencing Google's competitive push
2025
Gemini Scam Detection launches on Pixel devices in US, Australia, Canada, India, Ireland, UK
2026-02
Gemini replaces Google Assistant on Android with advanced on-device capabilities
📰

AI 週報

閱讀本週精選 AI 大事摘要 →

👉相關動態

AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源: IT之家

這是摘要,不是原文。去看原站,或訂閱每週簡報。

每週 AI 簡報

每週一封,可隨時退訂。