A former scientist from Fourth Paradigm ventures into the AI toy market and secures a seed - round financing of millions of US dollars from Amoeba Capital, SenseTime, and Fourth Paradigm | Exclusive Report by Hard Krypton

Author | Lin Qingqing

Editor | Peng Xiaoqiu

Yingke has learned that the AI hardware company "Artificial Productivity" has completed a seed - round financing of millions of US dollars. The investors include Amoeba Capital, SenseTime Guoxiang Capital, and Fourth Paradigm Fund. Yuanhe Capital will serve as the exclusive financial advisor. "Artificial Productivity" was founded in 2024 by Tu Weiwei, the former chief scientist of Fourth Paradigm. The first - generation AI companion robot, the Panda Robot, has been delivered to the outside world, and its first mass - produced smart toy for children is scheduled to be launched next month.

The "Artificial Productivity" Panda Robot at WAIC (Image source/Enterprise)

The global toy industry is currently facing structural challenges. According to data from the NPD Group, the market size of traditional toys was nearly one trillion yuan in 2024, but leading enterprises are facing growth pressure. At the same time, video games (such as NetEase's "Egg Party" with a peak daily active user count of over 50 million and over 70% of minor users) are rapidly occupying children's time. Parents are in a dilemma: traditional toys lack appeal due to their single interaction mode (such as the limited functions of remote - controlled cars), while electronic devices pose health risks such as vision damage and distraction.

"Artificial Productivity" is trying to use large - model technology to solve the growth dilemma of the traditional toy industry. "This generation of large - model technology has brought a revolutionary change in human - machine interaction, enabling machines to interact with humans in a more natural way. This open - style interaction revolution will free toys from the shackles of traditional 'remote controls' and endow them with richer connotations," Tu Weiwei, the founder of "Artificial Productivity", told Yingke. "However, merely having natural interaction with large models cannot solve the problem that toys are 'not fun'."

During early tests in children's shopping malls, the team found that although the Panda Robot of "Artificial Productivity" has dozens of functions, the two most inconspicuous functions - rock - paper - scissors and action imitation - are the most popular. They can even attract some children to stay for more than two hours, and the overall replay rate exceeds 80%. This fully verifies the market acceptance of "fun" functions rather than large - model conversations themselves. This highlights the flaws of current so - called "AI toys" - most products are actually plush toys embedded with general voice modules, which are just conversational storytellers and lack real fun. The "one - size - fits - all" large - model shells are not fun.

"We need to use AI toys to make children feel happy in a more open and intelligent way. Fun is the essence of toys," Tu Weiwei pointed out. "Every child is different. We need to learn from the closed - loop of user behavior feedback the personalized and unique ways for AI toys to bring happiness to each child."

In terms of core technology, the autonomous decision - making ability of AI Agents is the key to achieving "personalized and unique happiness". "Artificial Productivity" has independently developed the "Artificial Productivity" Autonomous Hardware Middle Platform (AP Agentic Hardware Platform), which includes three core capabilities: First, the "brain" of the AI Agent, which has multi - modal interaction and perception capabilities (perceiving user and toy body behaviors through multiple sensors), user behavior prediction capabilities (predicting user states and behaviors), decision - making capabilities (learning from the closed - loop data of behavior feedback to make decisions centered on user experience), and execution capabilities (ensuring natural and smooth actions based on the reinforcement learning control strategy of the world model). Second, the ability to optimize software - hardware collaboration, aiming to ensure the full release of AI capabilities while maximizing the cost - performance ratio of products. Finally, personalized interactive content, that is, using AIGC technology to co - create content with users to provide richer personalized experiences for AI hardware.

This architecture makes the hardware fully fun, and its pricing also has an obvious advantage compared with market competitors. The prices of some models even drop to the double - digit range. In actual tests, the system can dynamically adjust the decisions of the AI toy Agent according to children's emotions. For example, it can detect the small details of a child cheating during rock - paper - scissors and handle each child's small situations in a personalized way. Another example is generating "dramatic losses" during a winning streak to maintain the child's sense of participation.

"Creating a hit product is not a matter of luck. The supply side needs to quickly iterate and test for errors to continuously meet the ever - changing needs of the consumer market," Tu Weiwei pointed out. This middle - platform ability enables "Artificial Productivity" to quickly test for errors at a very low cost. The iterative experience of each product will be precipitated in the middle platform, which will continuously accelerate the iteration of new products in the future to meet the needs of people of different forms and age groups.

In terms of product form, "Artificial Productivity" has chosen to cooperate with top - tier IPs at this stage. It implants IP content into the AI toy Agent system and can also use props to trigger the advancement of the game process. The natural interaction interface allows a single piece of hardware to carry dozens of colorful gameplay modes. On the supply - chain side, "Artificial Productivity" cooperates with well - known international toy manufacturers to optimize the product's appearance design and structure and control costs. At the same time, by independently developing electronic modules, it provides users with high - cost - performance options at a significantly lower cost than similar products on the market.

In addition, in terms of channels, different from most AI toys on the market that focus on online sales, "Artificial Productivity" has started cooperation with most of the top online and offline channels on the market and will fully distribute products through e - commerce, Douyin, new retail systems, department stores, etc. "Online and offline channels each have their own advantages. Parents like to shop online but also like to take their babies to shopping malls. Toys need a good combination of online and offline channels to more efficiently reach more consumers."

In terms of the team, the founder, Tu Weiwei, is a doctor from the School of Artificial Intelligence at Nanjing University, the former chief scientist of Fourth Paradigm, and the former architect of the Baidu Fengchao Distributed Learning System. He has rich experience in AI technology R & D and cross - industry application implementation. The core R & D team comes from top universities such as Tsinghua University, Peking University, Fudan University, and Nanjing University. They have served leading enterprises in the industry such as Huawei, Alibaba, ByteDance, Kuaishou, Xiaohongshu, and Fourth Paradigm and have rich experience in software and hardware development and implementation.