Published Date : 16/08/2025
Alibaba is making a significant move into the smart glasses market with the introduction of the Quark AI Glasses. This device, powered by Alibaba's Qwen large language model and its AI assistant, Quark, represents the company’s first foray into the wearables category. The glasses are part of a broader $52.4 billion investment in AI and cloud computing and are scheduled to launch in China by the end of 2025.
The Quark AI Glasses will offer a variety of features, including hands-free calling, music streaming, real-time translation, meeting transcription, and a built-in camera. These capabilities are designed to enhance user convenience and productivity. Moreover, the glasses will integrate seamlessly with Alibaba's extensive ecosystem, allowing users to access navigation, make payments through Alipay, compare prices on Taobao, and tap into other Alibaba-owned platforms like mapping and travel booking.
Alibaba, based in Hangzhou, has been an active player in China’s AI development, rolling out models to compete with those from companies like OpenAI. By entering the smart glasses market, Alibaba joins a growing group of tech players betting on wearables as the next major computing platform alongside smartphones. The market already includes competitors like Meta’s smart glasses made with Ray-Ban and Xiaomi’s recently launched model.
Smart glasses like Alibaba’s depend on advanced AI systems that can recognize images, interpret context, and respond in natural language. These abilities are built on vast amounts of labelled data, which has been reviewed and tagged by humans to train the AI. The process often involves “human-in-the-loop” (HITL) systems, where people provide input at key stages of training and testing.
To gain insight into how HITL works, AI News spoke with Henry Chen, co-founder of Sapien, a company that manages large, distributed workforces for data labelling. Chen clarified common misconceptions about HITL, emphasizing that it involves more than just data labelling. It includes making decisions on edge cases, providing judgment calls, and offering ongoing evaluation. “Continuous feedback is what makes HITL work instead of one-off datasets,” he said.
Another misconception is that HITL work is low-skilled. Chen noted that the rise of industry-specific AI has created demand for domain experts like doctors, lawyers, and scientists to contribute their knowledge. Sapien works with 1.8 million contributors in 110 countries, and maintaining quality for complex tasks is critical. The company uses peer validation, contributor reputation tracking, and aligned incentives to ensure consistent results.
China’s AI sector is rapidly expanding, and the demand for data labelling is catching up to the levels seen in the US. While China has its own rules and regulations, the types of projects are increasingly similar to those in other major markets. Sapien uses on-chain technology to make payments transparent and give the community a say in which projects are worth pursuing. By operating without traditional offices, the company avoids some workplace issues and focuses on rewarding contributors for the value they deliver.
Automation is changing data labelling, but Chen believes humans will remain central to certain types of work. Tasks involving cultural nuance, sarcasm, rare diseases, niche languages, or complex sentiment will still require human review. “Humans will shift focus towards long-tail data and new vertical domains,” he said, predicting a rise in AI-assisted labelling while people handle the most challenging cases.
Sensitive projects, such as the IP of large corporations or international organizations, require strict controls. Sapien vets and trains enterprise contributors, uses data minimization and access controls, and follows compliance rules set by clients. The company operates under frameworks like SOC 2 Type 2, GDPR, and HIPAA.
As AI models become better at learning from unlabelled data—known as self-supervised learning—some expect the need for human labelling to shrink. Chen sees the role of human contributors evolving rather than disappearing. “We will evolve into a more specialized industry,” he said, noting that Sapien is already doing more work on evaluating synthetic data and model outputs. He expects future projects to focus on curating unique “ground truth” datasets, assessing AI performance, and providing domain-specific expertise.
Alibaba’s smart glasses highlight how far AI has moved into everyday products. While they may be one of many wearable devices in the market by 2025, the combination of Alibaba’s in-house language model, its existing services, and hardware integration could make them stand out for users in China. These devices depend on a complex supply chain of human expertise, from the engineers building the models to the contributors refining the data they use. Companies like Sapien operate behind the scenes, ensuring AI systems have the information they need to function more accurately and responsibly.
Whether in the form of smart glasses, virtual assistants, or other yet-to-be-released devices, AI-driven hardware is becoming a new way for companies to bring their services directly to consumers. For Alibaba, the Quark AI Glasses are both a product launch and a statement about where it sees growth—in technology that combines software, hardware, and human input.
Q: What are the key features of Alibaba's Quark AI Glasses?
A: The Quark AI Glasses offer hands-free calling, music streaming, real-time translation, meeting transcription, and a built-in camera. They also integrate with Alibaba's ecosystem, allowing users to access navigation, make payments through Alipay, and compare prices on Taobao.
Q: When will the Quark AI Glasses be available?
A: The Quark AI Glasses are scheduled to launch in China by the end of 2025.
Q: What is the role of human-in-the-loop (HITL) in AI development?
A: HITL involves humans providing input at key stages of training and testing AI models. This includes making decisions on edge cases, offering judgment calls, and providing ongoing evaluation to ensure the AI learns and performs accurately.
Q: How does Sapien ensure the quality of data labelling?
A: Sapien uses peer validation, contributor reputation tracking, and aligned incentives to maintain high-quality data labelling. They also vet and train enterprise contributors and follow strict compliance rules set by clients.
Q: What are the future trends in AI data labelling?
A: While automation is changing data labelling, humans will remain central to tasks involving cultural nuance, sarcasm, rare diseases, niche languages, or complex sentiment. The focus will shift towards long-tail data and new vertical domains, with a rise in AI-assisted labelling.