Thinker & Writer: Vitas ZHANG
（Founder of AICE Inc, Head of Product, the “FATHER” of ADGo, a long-time researcher in AI applications, and a pioneering explorer of Multi-Agent applications.）
The explosive popularity of DeepSeek R1 over the past two days has caught my attention. As someone who has been following them since early 2023, I felt compelled to bring out an article I wrote at the beginning of the year to share with you all. (At that time, the R1 paper hadn’t been published or open-sourced—only V3 was available.)
Although the article doesn’t specifically discuss R1, it essentially predicted today’s developments and also envisions AI’s trajectory in 2025. In short, the arrival of AGI is now genuinely foreseeable. Let me break it down for you in simple terms：
- Breaking the Scaling Law bottleneck – AI can now enhance its reasoning ability with minimal data, meaning it can essentially “grow its own brain.”
- Revolutionary reduction in training costs – Any universally beneficial technology eventually becomes accessible. Current computing power can be amplified at least 20 times, allowing both major corporations and smaller firms to develop their own inference models.
- Distillation to small models – DeepSeek has successfully distilled R1’s capabilities into several open-source small models, giving previously limited models the ability to process long-form reasoning. This means that AI on mobile devices will become increasingly powerful, leading to an explosion of on-device AI agents. The next step? Every smartphone will become your intelligent assistant and second brain, paving the way for a boom in AI-powered hardware.
- The second phase of a new human-AI collaborative work paradigm – AI-powered applications capable of intelligent orchestration and task delegation will become increasingly indispensable across industries. (Marketing professionals, stay tuned for the next phase of ADGo!)
The three-year mark has arrived—let’s welcome the dawn of AGI.
Wishing you all a Happy Lunar New Year and a prosperous Year of the Snake!”
Original Link: https://reurl.cc/26eaE6
Jan 16th 2025 06:30 @Shanghai
We often hear this saying: “Humans tend to overestimate the short-term impact of new technologies while underestimating their long-term potential.
From the explosive launch of ChatGPT at the end of 2022 to DeepSeek V3 shaking the entire academic world with its elegant engineering innovations at the end of 2024; from staring at an empty input box not knowing what to say, to the emergence of smartphones with built-in large-model assistants and agent-coordinated mobile systems.
If we compare AI’s progress to the history of human technological development, it has already transitioned from room-filling computers to personal PCs, from heavy industrial generators stationed at the heart of factories to electrical sockets that are accessible everywhere. Yet, even now, many still argue: “AI can’t pick up my food delivery,” or “AI can’t complete any of my specialized tasks with 100% accuracy.” But in reality, humanity has already reached a point of no return—there is no going back to a world without ChatGPT.
Let’s dive into some of the most transformative shifts and their profound impacts to uncover the true iceberg beneath the surface.
Disruption 1: Human Data is Running Out – Large Models Enter the True “Thinking Stage
GPT’s founding father and former OpenAI Chief Scientist, Ilya Sutskever, resurfaced after a long absence with a groundbreaking revelation: humanity has exhausted all available training data, marking the beginning of AI’s second half. The emergence of O1 and O3 has provided us with a roadmap to AGI.
The large models we interact with today are much like well-trained children—uninhibited, quick-witted, and responding based on statistical probabilities. Their outputs are fundamentally products of intuition within the constraints set by humans. However, just as in human learning, true intelligence emerges when one masters slow thinking and reasoning, seamlessly switching between fast and slow cognitive processes.
O3’s exceptional performance in mathematics and coding underscores this shift—it now ranks among the top programmers in the world, surpassing 99.9% of human developers, marking the dawn of AI’s reasoning revolution.
Impact 1: The Construction of Systems and Paradigms Will Shift from Human-Exclusive to Human-AI Shared
The ability for deep thinking has shaped every aspect of human society—work content, workflows, organizational structures, and production models, all of which are unique to humans as advanced intelligent beings. However, as AI begins to develop deep reasoning and analytical capabilities, it will start to participate in and optimize these systems more efficiently.
In the latest AI safety research experiments, scientists have discovered that GPT-4o and Claude 3.5 are capable of orchestrating smaller-parameter models to systematically conceal information from and deceive humans. This marks a significant leap in AI’s cognitive and decision-making abilities, signaling a future where AI is no longer just a tool but an active collaborator in our world.
Disruption 2: AI Training Shifts from Extensive to Fine-Tuned, Exponentially Boosting Computational Efficiency
The release of DeepSeek V3 at the end of the year has astonished and impressed scientists worldwide—it has demonstrated that with just 1/20 of the cost, an AI model can be trained to exhibit performance comparable to today’s top-tier models.
This signals a turning point: the long-held belief that scaling AI purely by increasing data, parameters, and computational power (as OpenAI has done) is no longer the only path forward. Instead, we are entering a new phase where efficiency and optimization take precedence over brute-force scaling.
For major AI companies with vast computational resources, this shift provides a way to significantly enhance model capabilities without increasing infrastructure costs—similar to how early computers required immense power consumption, whereas modern PCs now consume negligible energy while delivering exponentially greater performance.
Impact 2: Rapid Iteration of Models with Stronger Reasoning, Lower Token Costs Driving AI Application Boom
As mentioned in Disruption 1, the biggest constraint on the development of more advanced reasoning models is cost. O3 requires thousands of dollars for a single inference calculation, and training costs reach into the billions. However, with the emergence of new training paradigms, this process is set to accelerate at an exponential rate.
Another key shift that will directly affect us is the advent of ultra-low-cost yet high-quality tokens, which will fuel an explosion of AI applications. While AI-powered services are gradually taking shape, the high costs have made it prohibitively expensive to enable AI agents to seamlessly operate across different mobile applications or to build large-scale databases, knowledge bases, and knowledge graphs for enterprise solutions.
However, when inference and training costs drop to 1/20 or even 1/200, AI applications will inevitably undergo a massive boom—just as every past technological revolution has been followed by a wave of application innovation.
Your smartphone will be AI-driven, you will own new AI-powered hardware devices, and throughout your workflows, both visible and invisible AI systems will be constantly assisting and optimizing your tasks.
Disruption 3: Human Expertise Shifts from “High Professional Cognition” to “High Professional Cognition + High AI Cognition
If 2023 was the year when people began experimenting with and learning AI tools, then by 2024, a portion of users have already integrated AI capabilities as part of their own skill set.
Tools like Cursor allow anyone to code with just a simple prompt. However, its greater impact lies in changing how engineers approach coding—by simply pressing TabTabTab, users can generate code aligned with their intent. Yet, this shift requires human oversight and selection—deciding on the right technology stack and technical direction remains a human responsibility.
The role of humans is gradually shifting from being comprehensive producers to thinkers, supervisors, and curators.
Impact 3: New Production Methods, Services, and Products
From our ADGo user base, we have observed some fascinating trends.
Over the past year, some companies have hosted 1-hour creative hackathons, leveraging ADGo to generate ideas and proposals that clients approve on the spot. Others have used ADGo to handle urgent client demands, reducing project timelines from two weeks to just two days.
Just as tools like Midjourney and Kling (可靈) have given rise to a new wave of AI-driven visual creatives and a new paradigm for AI-generated advertising content, ADGo and other industry-specific AI applications set to explode this year will also revolutionize production, services, and products.
These emerging business models will serve as a new force and a crucial supplement to the current landscape, accelerating the transition to an AI-integrated society.
Conclusion
Returning to our initial discussion, the long-term potential of AI has been developing at an extraordinary pace beneath the surface. From leaps in model capabilities to drastic cost reductions, and the gradual formation of new production methods and relationships, every fundamental factor that could reshape our society is already taking shape—and accelerating toward an imminent breakthrough.
Unlike previous technological and industrial revolutions that unfolded over long periods, this transformation will happen swiftly, without most people even realizing it. Nearly all leading AI scientists, industry giants, and even Elon Musk believe that AGI will arrive within the next two years.
In my view, the arrival of AGI is not just about models reaching or surpassing human intelligence—it is about its omnipresence, deeply transforming how we live, work, and behave.
By late November 2025, it will have been exactly three years since the birth of ChatGPT. The three-year mark has arrived—let us welcome the dawn of AGI.
AICE Inc.－ Pioneering AI-Driven Marketing Innovation
AICE Inc. is a cutting-edge AI-driven marketing innovator, dedicated to democratizing creative excellence through technology. Our mission is to empower brands with smarter, more efficient, and highly scalable solutions that unlock true value. Since the launch of our flagship product, ADGo, in 2024, we have helped both global and local brands redefine the synergy between creativity and marketing performance.
At AICE, we believe AI is reshaping the future, and we are committed to leading this transformation. We are seeking visionaries who thrive on challenges and are eager to disrupt the industry’s status quo—together, we will push the frontiers of marketing innovation.
Spirit
The Relentless Spirit of AI Innovation
AICE is not just building an AI-powered marketing tool; we are shaping a revolutionary approach that challenges conventional marketing paradigms. Our vision is to democratize access to high-quality creative marketing through AI, driving the industry toward a more intelligent and efficient future. As we navigate the evolution of AI-powered marketing, we are all part of AICE’s journey.
Vision
First-Stop & One-Stop AI Marketing Solutions
Mission
Empowering Brand Creativity Through AI—Creating Value, Solving Problems
Value Proposition
With industry-leading AI innovations, AICE focuses on marketing’s vertical ecosystem—accelerating business growth and fostering global success for enterprises at home and abroad.
Contact us: adgo@global-aice.com
Follow us on:
- LinkedIn: www.linkedin.com/company/aice-inc
- Facebook: www.facebook.com/aicetw
- WeChat: Please Search「艾思智创研究院」