Chinese Researchers Release Five-Stage Roadmap for Artificial Intelligence
Leading Chinese institutions, including ByteDance and Tsinghua University, have published a five-stage roadmap for artificial intelligence to achieve self-improvement.
Researchers from leading universities and technology companies in China have unveiled a new five-stage artificial intelligence roadmap aimed at creating systems capable of developing their own versions without human intervention.
Five-Stage Iterative Development Process
A paper jointly published by institutions such as ByteDance, Tsinghua University, and the Shanghai Artificial Intelligence Laboratory outlines a five-stage process for iterative self-improvement. In the first stage, the artificial intelligence executes procedures designed by engineers, while in subsequent stages, it begins to choose how to update itself.
Adaptation in Advanced Stages
It is envisioned that in advanced stages, the systems will adapt to post-deployment changes by determining what new knowledge or experience they need. In the final stage, the goal is for the artificial intelligence system to continuously refine the methods used to improve artificial intelligence.
Potential to Reduce Costs
It is stated that automating a portion of artificial intelligence research could provide a competitive advantage for model developers. This paradigm shift is expressed as having the potential to shorten development cycles while reducing the labor and computational costs required to build foundational models.
Competition with the US and Hardware Constraints
Automating artificial intelligence research constitutes a critical front in the technological competition between the US and China. Experts note that US firms hold an early advantage due to superior hardware resources, whereas Chinese researchers succeed in extracting high performance from scarce resources.
Corporate Moves Toward Self-Training
The company Z.ai announced that it will allocate a major portion of the revenue from its recent funding round to the development of next-generation foundations and fully self-training systems. Other organizations such as MiniMax and DeepSeek are also working on systems that can update their memories and provide greater autonomy.
Safety Measures and Domain Differences
Researchers noted that the five-stage process follows a clearer path in software engineering, whereas robotics and scientific discovery face steeper obstacles. It was emphasized that strict safety measures, such as rigorous testing environments to ensure updates are safe and beneficial, are essential for true iterative development.