A Chinese humanoid startup flips 'distillation' claim on OpenAI as it releases a new robotics model

The accusation by JoyIn's CEO, Guo Renjie, marks a significant turn in the ongoing debate surrounding intellectual property and AI development. Historically, U.S. companies like Anthropic have frequently accused Chinese firms of 'distillation' – essentially, unauthorized replication or learning from their advanced AI models. JoyIn's claim, however, flips this narrative, suggesting that a major U.S. player, OpenAI, may have engaged in similar practices.

Guo Renjie

Guo Renjie, the CEO of Suzhou-based JoyIn, publicly voiced his concerns in a letter to OpenAI, which he shared on Thursday. He explicitly stated, "People often say major tech companies have intelligence networks monitoring the whole internet, this time I believe it, this is a direct distillation of us without any modifications." This strong assertion underscores the perceived directness of the alleged copying, moving beyond mere inspiration to a claim of direct appropriation of technical concepts and design elements. Guo's statement suggests a belief that OpenAI's actions were not coincidental but rather a deliberate act of leveraging JoyIn's public presentations.

Guo highlighted specific technical similarities, including the use of "recursive self-improvement" and AI optimization of computing power, concepts he claims were part of JoyIn's "extraterrestrial visitor" AI model framework presented in Silicon Valley weeks before OpenAI's Chief Scientist published "An Alien Mind." Furthermore, he pointed to the outer space-inspired design of OpenAI's GPT-6 Astra web page, noting its resemblance to JoyIn's Aether model website, which he claims was released two months prior. These detailed accusations suggest a pattern of alleged imitation across both the underlying technology and its public presentation, leading JoyIn to initiate legal proceedings.

Aether Model

JoyIn's Aether model is central to this controversy, with Guo Renjie asserting its prior public release and unique capabilities. The model, developed under the leadership of Zhu Mingxuan, who previously worked on humanoid models at U.S. company Figure, focuses on a perceptive rather than text-based approach to robotic control. This design choice is claimed to enable humanoids to achieve a 90% success rate on tasks on the first attempt, a significant benchmark in robotics. Zhu Mingxuan also stated that Aether reduced training time by two-thirds, indicating a potentially efficient and advanced system.

The company's decision to open-source parts of the model, such as those related to touch, while retaining control over energy use components, reflects a strategic approach to collaboration and proprietary protection. This selective open-sourcing could be seen as an effort to contribute to the broader AI community while safeguarding core competitive advantages. The Aether model's capabilities and its development timeline are crucial to substantiating JoyIn's claims against OpenAI, as the timing of its public presentation relative to OpenAI's releases forms the basis of the alleged distillation.

Y Combinator

The broader context of AI development and intellectual property is further complicated by comments from industry figures like Garry Tan, CEO of Y Combinator. Tan's stance, stating he would "do nothing" about distillation and pointing out that U.S. companies have trained their AI models on copyrighted data, introduces a layer of complexity to the ethical and legal landscape. His remarks suggest a prevailing view within some parts of the tech industry that the lines around intellectual property in AI are already blurred, with many foundational models potentially built on data that raises its own copyright questions.

This perspective could influence how seriously claims of distillation are taken, particularly when they involve cross-border accusations. If leading figures acknowledge that AI training often involves data with ambiguous copyright status, it might create a precedent that makes it harder to prosecute or even define 'distillation' clearly. The ongoing tech race between U.S. and Chinese companies, coupled with these complex IP issues, underscores the need for clearer international guidelines and legal frameworks to govern AI development and protect innovation.