内容来源：https://lifehacker.com/tech/meta-muse-sparks-ai?utm_medium=RSS

内容总结：

Meta发布新一代AI模型Muse Spark，迈向“超级智能”新阶段

近日，Meta公司正式推出全新升级的AI模型——Muse Spark，并将其定义为迈向“超级智能”的重要一步。该模型被描述为“原生多模态推理模型”，其功能远超传统聊天机器人，未来将深度集成至智能眼镜及社交媒体信息流中。目前，Muse Spark已上线Meta AI应用程序，并计划于未来几周内通过固件更新登陆Meta与雷朋、Oakley合作推出的智能眼镜。

与以往“一刀切”的AI交互模式不同，Muse Spark引入了三级“思考”深度，用户可根据需求自主选择响应模式：

• 即时模式：适用于快速问答及日常对话。
• 思考模式：专为处理数学、科学、逻辑等复杂问题设计。
• 沉思模式：最高级别模式，可调动多个AI智能体并行协作，完成多步骤的复杂任务。

Meta宣称，Muse Spark在性能上媲美或超越了其Llama 4 Maverick模型，而计算资源消耗却降低了一个数量级以上，这意味着高效能推理有望在不过度依赖服务器的情况下实现。

多场景深度融合，智能眼镜成为核心载体
Muse Spark的一大突破在于其跨工具整合视觉信息的能力。例如，用户佩戴智能眼镜扫描杂乱的电线与电子设备后，可直接提问“如何连接这套家庭影院系统？”；在组装家具时，AI可实时识别步骤并指导操作，避免装错零件。该模型深度结合视觉素材的设计，尤其适合智能眼镜这一穿戴场景。

健康推理功能获医学支持
Meta透露，其超级智能实验室与逾千名医生合作，为Muse Spark开发了健康推理能力。未来用户可通过AI解析食物营养成分，或在健身时获取肌肉激活状态的可视化指导。

实际效能仍待检验
尽管技术演示令人印象深刻——在测试中，Muse Spark成功识别了音频设备图片，并提供了准确的连接方案与所需线材——但AI技术在真实场景中的表现仍是关键挑战。复杂光线、网络延迟及实际任务的多样性，都将考验其落地可靠性。

目前，用户已可通过Meta AI官网及应用程序体验Muse Spark，智能眼镜固件更新与社交媒体集成预计将在短期内陆续推出。

中文翻译：

Meta宣布推出全面升级的AI系统Muse Spark，称其为迈向"超级智能"的重要一步。这款"原生多模态推理模型"远不止于聊天机器人，即将融入智能眼镜与社交信息流。目前该系统已在Meta AI应用程序上线，并计划在未来数周内通过智能眼镜固件更新向用户推送。

Muse Spark采用三级"思考"模式，摒弃了传统的一刀切方案，用户可自主调节AI的思考深度：
即时模式：适用于快速问答与日常对话
思考模式：专为数学、科学及逻辑等复杂问题设计
深度推演模式：最高级别模式，可调动多个并行工作的AI智能体协同完成多步骤复杂任务

Meta表示，Muse Spark在性能上媲美甚至超越Llama 4 Maverick模型，而计算资源消耗却降低了一个数量级。这意味着理论上系统可实现高效推理，同时大幅减少服务器负载。

尽管Muse Spark将适配多种场景，但其底层视觉素材整合能力似乎专为智能眼镜量身打造。雷朋Meta和欧克利Meta智能眼镜用户将能体验以下新功能：

跨工具视觉信息整合
相较于前代模型，Muse Spark的核心突破在于跨工具视觉信息整合能力。例如，用户可对准杂乱的线缆与电子设备询问："如何连接这套家庭影院系统？"或在组装宜家家具时，无需翻阅说明书即可获得逐步指导。AI将自动解析安装步骤，确保用户不会装错零件。

健康推理功能
Meta超级智能实验室联合千余名医疗专家开发了健康推理功能。用户可生成交互式营养分析图表，或实时解析健身动作的肌肉激活状态。

实际表现究竟如何？
上述功能目前仍属"理论阶段"。人工智能的实际表现往往难以匹配宣传声势——在实验室基准测试中表现出色是一回事，但在现实场景中，面对光线不足、网络延迟、复杂说明书等变量，才是真正的技术考验。

笔者虽未深入测试，但通过开启"思考"模式上传随机音频设备照片进行了快速体验：系统不仅准确识别了图中所有设备，还提供了多种连接方案，并正确列出了所需线材类型。这让人对眼镜端应用充满期待。

若想亲自体验，Muse Spark已在meta.ai网站及Meta AI应用程序开放测试，智能眼镜固件更新与社交媒体集成预计将很快跟进。

英文来源：

Calling it a step towards "super intelligence," Meta announced it is releasing Muse Spark, an overhauled and improved AI. This "natively multimodal reasoning model" goes way beyond a chatbot, and it will soon live in your glasses and your social feeds. It's available now in the Meta AI app, with plans to roll out with a smart glasses update in the next few weeks.
Instead of a one-size-fits-all approach, there are three levels to Muse Spark's "thinking," and users will be able control how deep the intelligence goes.
Instant Mode: For quick questions and everyday chats.
Thinking Mode: This mode is designed to solve more complex problems, so if you need some help with math, science, or logic, this is the mode.
Contemplating Mode: Muse Sparks' highest level engages multiple AI agents that work in parallel and collaborate to complete complex, multi-step tasks.
Meta says Muse Spark's performance compares to or exceeds their Llama 4 Maverick model while using over an order of magnitude less computing power. That means, theoretically, high-level reasoning without excessive server use.
While Muse Sparks will be accessible in a variety of places, it seems like Muse Spark's ground-up integration of visual material is made for smart glasses. Here are some of the ways Ray-Ban Meta and Oakley Meta users will be able to use the new AI.
AI is now integrated across different tools
One of the Muse Spark main improvements over Meta's previous model is the way the new AI will integrate visual information across different tools. So, theoretically, you could point your glasses at a mess of wires and electronic boxes and say "how do I hook up this home theater system?" Or get step-by-step coaching on assembling a piece of IKEA furniture without opening the booklet. The AI would read the instructions and make sure you're not screwing anything in upside down.
Muse Sparks will have health reasoning capabilities
Meta said its Meta Superintelligence Lab collaborated with over 1,000 physicians to develop the AI's health reasoning capabilities. Users will be able to do things like generate an interactive display that unpacks the nutritional information about food, and maps out what muscles are activated during a workout.
But how will it actually perform?
All of the above is "in theory." Artificial intelligence hasn't always lived up to its hype, even when it's being hyped in front of a massive audience. It's one thing to perform well in laboratory benchmark tests, but how the tech works in the real world, where lighting is spotty, wi-fi is slow, and furniture instructions can be extremely complicated, is the real challenge.
While I haven't dug deeply into the tech, I did give it a quick test by turning on "thinking" mode and sending Meta AI the below picture of a random assortment of audio gear:
It not only correctly identified everything in the picture, it gave me a couple different options for possible ways to hook it together, and told me (correctly) what cords I needs. So I look forward to having it on my glasses. If you want to test it yourself, Muse Spark is already running on meta.ai and the Meta AI app, and smart glasses firmware and social media integrations are expected to follow shortly.