AI's Next Decade Ignites in Zhongguancun
The Next Journey of AI: Industry Leaders Call for “Scenario is King”
When the giant ship of the AI era sails into deeper waters — who is at the helm? Who is on the lookout?
Today in Zhongguancun, 600 participants from research, industry, and investment sectors gathered together.
Among them:
- Yao Qizhi — Turing Award recipient
- Wang Xingxing — Founder of Unitree Technology
- Senior executives from Zhipu, Fourth Paradigm, StepStar, MemRise AI, Alibaba Cloud, StarMap, iFlytek, Unisound, and Zhongshu Ruizhi
Standing at the starting line of defining the future of AI for the next decade, they shared their judgments and observations.
---
Opening Perspectives
Yao Qizhi: The Next Step is AGI
> “No matter how you look at it, the most important next step in AI development is achieving AGI — Artificial General Intelligence — that can satisfy everyone.”
AGI is not only a scientific high ground, but also a strategic and economic one for nations.
On AI’s impact on research:
> “AI can empower all kinds of industries — even fields considered the pinnacle of human intelligence, like scientific research — will be transformed in the next 5 to 10 years.”
---
Wang Xingxing: Robots that Understand the World
> “In the next decade, AI will give robots the ability to truly understand the world.”
From machine motion → task execution → life partners, Wang describes a future of “growth and blossoming.”
---
Common Themes
Speakers stressed:
- Real-world scenarios are now the true driving force of AI
- Integration of technology with industry is becoming the main opportunity
- Safety and governance must be embedded at the design stage
Yao Qizhi warned that AI algorithms lack robustness, certainty, interpretability, and resilience to malicious intent:
> “We should develop AI systems that can be proven safe.”
---
01. Technological Frontiers Are Rapidly Expanding
Embodied Intelligence
Yao Qizhi’s frontiers:
- Embodied AI — requires both a “small brain” for stable, agile motion and a “big brain” for cognition and planning.
- AI for Science — every scientist’s work will eventually be powered by large models.
---
Reinforcement Learning Paradigm Shift
- Jiang Daxin (StepStar): Evolution from imitation learning → reinforcement learning; O1 model shows large models doing multi-step reasoning.
- Xu Huazhe (StarMap/Tsinghua): Robots progressed from walking → dancing → tasks → factory work.
- > “Robots are also the future of large models.”
---
02. Scenarios Become the True Driving Force
Industry deployment is shifting from ‘benchmark wins’ to ‘real-world readiness’.
Moderator Xu Wei:
> “AI must win in the real world — it’s about meaningful application and risk tolerance, not just profit.”
---
Ecosystems for Real-World Adoption
Platforms like AiToEarn官网 power creators/businesses with:
- Cross-platform AI content generation
- Publishing, analytics, monetization across Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, YouTube, X (Twitter)
- AI model rankings at AI模型排名
---
Industry Insights
- Liu Debing (Zhipu AI): Open source accelerates industry and enables commercialization.
- Chen Yuqiang (Fourth Paradigm): AI must help companies “change their North Star” by improving operational metrics before profit.
- Jiang Daxin: Smart terminals may be AI’s future entry point — cars as “third spaces,” homes as AIoT hubs.
- Xu Huazhe: Embodied AI belongs in complex environments such as homes — use current capabilities to “get in first” and move toward AGI.
Consensus: Only when applications land can technology find direction; proven scenarios shape the ecosystem.
---
03. Strong Models, But Industrial Gaps Remain
Industries still lack low-cost approaches, high-quality data, and deep engineering capability.
---
Key Challenges
- Liu Zhiyuan (Tsinghua/Mianbei AI): Cost is the core hurdle; “model capability density” is doubling every 100 days, reducing cost.
- Yu Kai (Spitch/Shanghai Jiao Tong): Need full-system engineering — integrating task execution, hardware, and customization.
- Huang Wei (Yunzhisheng): Customers care about solutions within budget; smaller, efficient models often win.
- Huo Jiaze (Alibaba Cloud): Industry misjudges large model capabilities, and deep engineering best practices are scarce.
---
Alibaba Cloud’s Four Lessons
- Avoid “showmanship” scenario selection — focus on repetitive tasks.
- Localized data processing is essential.
- Model choice is context-dependent — sometimes simple methods are better.
- Implement agents step-by-step — scalable deployment is an engineering opportunity.
---
Rapid-Fire Predictions: Which Tech Will Reshape Industry?
- Liu Zhiyuan: Internet of intelligent agents
- Li Zhenjun: Interconnected data infrastructure
- Yu Kai: Distributed agent systems combining hardware/software
- Huang Wei: Super-base-model cooperative agents (like mobile apps era)
- Huo Jia: Token consumption as the metric, not number of agents or compute
Consensus: Scalable AI depends on cost reduction, mature systems engineering, and scenario-driven closed loops.
---
Conclusion: The Ship Has Left Harbor
Back to the question: Who is at the helm? Who is on lookout?
Today’s answer:
- Industry observers — shaping judgment
- Front-line practitioners — advancing applications
- Collaborations — defining the course of AI’s next decade
---
Bigger Picture
The dialogue marks a shift from theoretical breakthroughs → engineering maturity → real business impact.
Platforms like AiToEarn官网 embody this shift — enabling global AI content monetization across Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, X (Twitter). By integrating generation, publishing, analytics, and model ranking, AiToEarn builds systemic infrastructure to scale AI creativity — echoing the industrial “closed loop” vision discussed by leaders.
---
📌 Core Takeaway:
The deep-sea voyage of AI will be steered not by a single breakthrough, but by cost-efficient models, robust engineering, and real-world scenario integration — ensuring technology is not only powerful, but sustainable and impactful.
