|
The ads include videos of people doing housework for money. |
Recently, some social media users in Vietnam saw advertisements for a strange mobile app. The person in the video encourages users to download the app and record themselves doing housework to earn money. The ad describes that simply recording while doing dishes, laundry, etc. can generate additional income. According to the ad, the user data sold will be used to train the AI.
This app is currently available on Android and iOS and allows easy registration. However, the company that supports it has strict standards when it comes to operation. It requires partners to wear their phones on their heads and place their hands in designated areas while operating the app. In addition, according to reviews from previous users, the company pays a low fee per minute of video. It also often rejects uploaded material that is shaky, blurry, or otherwise ineffective.
 |
|
The app collects a lot of data and claims ownership of uploaded clips. |
By registering, users become partners of the platform. Account holders must agree to a long list of terms and conditions, which stipulate that the data collected will become the property of the company behind it. In addition, they have the right to sell them to partners. This poses significant risks of leaking sensitive information.
Actually, getting paid to collect image data for AI and robot training is nothing new. However, not many platforms in Vietnam target human resources. Robotics companies need millions of hours of first-person video to teach “natural AI” to hardware devices.
Housework, such as washing dishes and sweeping, are basic activities for humans. However, they are complex for robots because they require the machines to learn how to coordinate hand strength, perceive soap slippage, grasp objects at right angles, and handle dropped objects. Robots must calculate impact level, force direction and how to interact with objects in a real-world environment.
Text-to-video models such as OpenAI’s Sora or Google Veo require huge, diverse, and proprietary video datasets. They require a deep understanding of real-world physics and the ability to physically reproduce motions.
Source:
