[Preview of the Weekly] As AI becomes more prevalent in end-user devices, how will it reshape the future of smartphones and computers? (AI Translation)
Listen to the full version
- Edge AI, which has been quietly empowering fields like photography, security, and connectivity, is now stepping into the spotlight as a new selling point for smartphone and computer manufacturers. Vivo showcased its AI assistant 'Jovi' at a developer conference, demonstrating its ability to quickly retrieve photos with complex queries. The company's self-developed Blue Whale model matrix supports this functionality, with a 7 billion parameter version already deployed on devices and a 13 billion parameter model running on the edge side.
- Major Android phone brands such as Samsung, OPPO, Xiaomi, Honor, and Huawei are joining the race in edge-side large models. Chipmakers like Qualcomm and MediaTek are also intensifying competition in providing underlying computational power for these models. Qualcomm's third-generation Snapdragon 8 is the first mobile chip platform to support decabillion-parameter models on the device side.
- While edge AI offers privacy benefits by processing data locally without sending it to the cloud, challenges remain in terms of processor power requirements and significant memory and storage space consumption. This could potentially lead to higher costs for consumers. Additionally, battery life poses another challenge due to high energy consumption by large models. Currently, edge-side large models are primarily used in intelligent assistants with functionalities limited to summarization, information retrieval, voice recognition, text translation, image and text content generation; no 'killer applications' have emerged yet.
相较于依托数据中心、通过处理巨量数据给出洞察建议的云端AI（人工智能），终端侧AI多在拍照、安全、联接等领域扮演并不显山露水的赋能角色。如今，它正在走向台前，成为手机、电脑厂商的新卖点。
Compared to cloud-based AI (Artificial Intelligence), which relies on data centers and processes vast amounts of data to provide insights and recommendations, edge AI often plays a less conspicuous but empowering role in areas such as photography, security, and connectivity. Today, it is stepping into the spotlight, becoming a new selling point for smartphone and computer manufacturers.
“找一下在雨天套垃圾袋大笑的女孩照片。”11月1日，vivo开发者大会上，该公司AI解决方案中心总监谢伟钦在一部手机上敲出这句话，照片几乎立刻现于屏幕。他称，几年前和妻子逛街时突遇暴风雨，伞被吹坏，只好向路边保洁人员讨两个垃圾袋挡雨。看着妻子在雨中狼狈又有些好笑的样子，他抓拍下这张照片。因记不清具体时间，原本每次想看都要找很久。
"Find the photo of the girl laughing in the rain with a garbage bag on." On November 1st, at the vivo developer conference, Xie Weiqin, director of the company's AI Solution Center, typed this sentence on a smartphone and the photo appeared on the screen almost immediately. He recounted that a few years ago, while shopping with his wife, they were caught in a sudden storm. Their umbrella was broken by the wind, so they had to ask sanitation workers for two garbage bags to shield themselves from the rain. Seeing his wife in an awkward yet amusing state in the rain, he captured that moment with a photo. Since he couldn't remember the exact time it was taken, he used to spend a long time searching for it whenever he wanted to revisit that memory.
他在展示的是vivo大模型加持下的手机助手“蓝心小V”。除了找照片、文件，它还能为闹市照片“一键消除”路人，阅读论文提炼重点，按关键词或图片写作社交媒体文案，以及头脑风暴并生成思维导图。vivo未公布这些功能哪些完全基于终端，不过其称，为蓝心小V提供支撑的自研蓝心大模型矩阵中，70亿参数版本已落地终端，130亿参数模型也已实现终端侧跑通。
He is demonstrating the mobile assistant "Lanxin Xiao V," enhanced by vivo's large model technology. In addition to locating photos and files, it can also "one-click remove" passersby from busy street photos, summarize key points from academic papers, write social media copy based on keywords or images, as well as brainstorm and generate mind maps. Vivo has not disclosed which of these features are entirely based on the terminal; however, they have stated that within the self-developed Lanxin large model matrix that supports Lanxin Xiao V, a version with 7 billion parameters has been implemented on devices, and a model with 13 billion parameters has also been successfully run on the terminal side.
Sign in with Google
Sign in with Facebook
Sign in with 财新