Kimi Yang Zhilin: Large Model Training Enters Third Stage of AI-Driven Research
Moonshot founder Yang Zhilin stated at the Zhongguancun Forum annual meeting on March 25, 2026, that large model training is entering a third critical stage driven by AI. This paradigm shift signals a move from relying on natural data and manual annotation toward highly automated self-evolution.

Reflecting on the technical roadmap, Yang Zhilin outlined three phases in large model evolution: the first phase three years ago primarily depended on natural internet data and limited manually annotated value alignment; the second phase last year centered on large-scale reinforcement learning, with researchers curating high-quality tasks to enhance model performance. In 2026, a fundamental shift has occurred in AI research methods, and the role of researchers is evolving into an "AI compute scheduler." In this new stage, the research process is driven by AI, which uses vast numbers of tokens to autonomously synthesize new tasks and environments, define optimal reward parameters, and even actively engage in exploring new network architectures.
This trend suggests that AI research and development efficiency is poised for exponential acceleration. Moonshot announced that its core product Kimi will focus on advancing the frontiers of intelligent technology and fostering a collaborative, evolving technology ecosystem with the open-source community. The transition from "human teaching AI" to "AI guiding research" is not only an upgrade in training methods but also a significant milestone on the path to achieving general artificial intelligence (AGI), signaling a shift from passive learning to autonomous exploration.
Related article
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
AI race won’t wait, nor will this TechCrunch Disrupt 25% discount
Every founder building in AI right now is asking the same question: Are we moving fast enough, and effectively enough? TechCrunch Disrupt 2026 puts you in the same room as the people wrestling with that existential question at the highest levels. Fro
Related Special Topic Recommendations
Comments (0)
0/500

Reflecting on the technical roadmap, Yang Zhilin outlined three phases in large model evolution: the first phase three years ago primarily depended on natural internet data and limited manually annotated value alignment; the second phase last year centered on large-scale reinforcement learning, with researchers curating high-quality tasks to enhance model performance. In 2026, a fundamental shift has occurred in AI research methods, and the role of researchers is evolving into an "AI compute scheduler." In this new stage, the research process is driven by AI, which uses vast numbers of tokens to autonomously synthesize new tasks and environments, define optimal reward parameters, and even actively engage in exploring new network architectures.
This trend suggests that AI research and development efficiency is poised for exponential acceleration. Moonshot announced that its core product
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
AI race won’t wait, nor will this TechCrunch Disrupt 25% discount
Every founder building in AI right now is asking the same question: Are we moving fast enough, and effectively enough? TechCrunch Disrupt 2026 puts you in the same room as the people wrestling with that existential question at the highest levels. Fro





Home






