更可靠的方向可能是混合架构:让 AI 负责理解自然语言和设计自动化,再由经过验证的确定性软件执行最终控制。
Why, roughly a year after Amazon and Google introduced Alexa Plus and Gemini for Home, do generative-AI smart-home assistants remain unreliaGenerative AI can make smart-home assistants more conversational, but dependable device control still requires deterministic execution.
AI 提示
Create a landscape editorial hero image for this Studio Global article: Why, roughly a year after Amazon and Google introduced Alexa Plus and Gemini for Home, do generative-AI smart-home assistants remain unrelia. Article summary: The core problem is that a smart-home controller must be predictable, fast, and correct every time, while a generative model is optimized to produce plausible, flexible language. Conversational fluency can improve intent. Topic tags: general, news, general web, user generated, documentation. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, water
openai.com
生成式 AI 让智能家居助手的表达方式更自然,却没有同步让它们变得可靠。这个落差在最简单的指令上最明显:开灯、把亮度调到某个百分比,或者启动一套例程。用户并不需要一个听起来合理的回答,而是希望正确的设备立刻切换到正确状态。
失灵的,恰恰是最基础的指令
问题并不只出现在复杂的多步骤自动化中。科技评论员 David Pogue 表示,他向 Alexa+ 发出 135 个请求,答对的“略低于一半”,其中包括开灯和设置调光亮度等操作。 这只是个人测试,并非标准化基准测试,但对于一款核心职责就是控制设备的助手来说,这一结果仍然相当醒目。
Google Home 用户也报告了一个更具体的故障:手机未连接 Android Auto 时,Gemini 可以正常启动 Home 例程;一旦 Android Auto 处于连接状态,同样的例程就可能无法运行。有用户称,助手只会给出随机或泛泛的回答。
更可靠的方向可能是混合架构:让 AI 负责理解自然语言和设计自动化,再由经过验证的确定性软件执行最终控制。
Why, roughly a year after Amazon and Google introduced Alexa Plus and Gemini for Home, do generative-AI smart-home assistants remain unreliaGenerative AI can make smart-home assistants more conversational, but dependable device control still requires deterministic execution.
AI 提示
Create a landscape editorial hero image for this Studio Global article: Why, roughly a year after Amazon and Google introduced Alexa Plus and Gemini for Home, do generative-AI smart-home assistants remain unrelia. Article summary: The core problem is that a smart-home controller must be predictable, fast, and correct every time, while a generative model is optimized to produce plausible, flexible language. Conversational fluency can improve intent. Topic tags: general, news, general web, user generated, documentation. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, water
openai.com
生成式 AI 让智能家居助手的表达方式更自然,却没有同步让它们变得可靠。这个落差在最简单的指令上最明显:开灯、把亮度调到某个百分比,或者启动一套例程。用户并不需要一个听起来合理的回答,而是希望正确的设备立刻切换到正确状态。
失灵的,恰恰是最基础的指令
问题并不只出现在复杂的多步骤自动化中。科技评论员 David Pogue 表示,他向 Alexa+ 发出 135 个请求,答对的“略低于一半”,其中包括开灯和设置调光亮度等操作。 这只是个人测试,并非标准化基准测试,但对于一款核心职责就是控制设备的助手来说,这一结果仍然相当醒目。
Google Home 用户也报告了一个更具体的故障:手机未连接 Android Auto 时,Gemini 可以正常启动 Home 例程;一旦 Android Auto 处于连接状态,同样的例程就可能无法运行。有用户称,助手只会给出随机或泛泛的回答。