JoyIn's CEO Guo Renjie has accused OpenAI of 'distillation' in its latest model, citing striking similarities and a 90% success rate in JoyIn's own Aether model. The Ant-backed startup has begun legal proceedings, though OpenAI has not yet responded.
TL;DR
- JoyIn's CEO publicly accuses OpenAI of 'distillation' in GPT-6 Astra, citing similarities in technology and design.
- JoyIn's Aether model claims a 90% first-attempt success rate for humanoid tasks, with plans to open-source parts of the model.
- The dispute highlights tensions in AI development between U.S. and Chinese companies, with broader implications for intellectual property and model training practices.
What happened
In a public statement, JoyIn CEO Guo Renjie accused OpenAI of directly 'distilling' JoyIn's Aether model into its own GPT-6 Astra. Guo pointed to similarities in core technological approaches, such as recursive self-improvement and AI-optimized computing power, as well as design elements inspired by outer space.
JoyIn's Aether model, announced publicly by Guo on Thursday, claims a 90% success rate for humanoid robots completing tasks on the first attempt. The model's perceptive approach to robotic control, rather than text-based, aims to outperform existing systems.
Guo mentioned that JoyIn has begun the process of filing a lawsuit, though CNBC was unable to independently verify the claims. OpenAI has not yet responded to requests for comment.
Why it matters
This dispute underscores the growing tensions in AI development, particularly between U.S. and Chinese companies. The accusation of 'distillation'—where one company uses another's model to improve its own—has been a point of contention, with U.S.-based Anthropic previously flagging similar issues.
For developers and startups, this case highlights the importance of intellectual property protections and the potential risks of open-sourcing parts of AI models. It also raises questions about the ethical and legal boundaries of model training and improvement.
Investors should watch this space closely, as the outcome could influence AI development strategies and partnerships. The broader implications for the tech race between the U.S. and China add another layer of complexity to an already competitive landscape.
Key facts
- JoyIn's Aether model claims a 90% first-attempt success rate for humanoid tasks, according to the company.
- Guo Renjie, CEO of JoyIn, accused OpenAI of 'distillation' in a public statement on Thursday.
- OpenAI's GPT-6 Astra was launched on September 3, while JoyIn announced its Aether model on September 10.
- JoyIn plans to open-source parts of the Aether model, excluding portions related to energy use.
- Zhu Mingxuan, who led the development of Aether, previously worked at U.S. humanoid company Figure.
- The Aether model reduces training time by two-thirds, according to Zhu Mingxuan.
- JoyIn has begun the process of filing a lawsuit, though details remain unverified.
- OpenAI has not yet responded to requests for comment on the accusations.
Context
The accusation of 'distillation' comes amid a broader tech race between U.S. and Chinese companies in AI development. U.S.-based Anthropic has previously raised concerns about unauthorized distillation of its models by Chinese companies.
This dispute also highlights the challenges in getting humanoids and AI models to perform as intelligently as humans across a variety of tasks. The success rate claimed by JoyIn's Aether model, if verified, could be a significant milestone in the field.
The open-sourcing of parts of the Aether model, while excluding energy-related portions, could influence the direction of AI research and development. It also raises questions about the balance between collaboration and competition in the AI industry.
