The process from “human-machine collaboration” to “human-machine symbiosis” is long, yet it is the most imaginative…
Author丨Bu Han
Editor丨Liu Huan
In just a few years, generative artificial intelligence (AIGC) has rapidly risen from an obscure academic concept to a hot topic of discussion among the public.
During the Spring Festival, Deepseek’s global explosion once again brought large models into the spotlight.
At least from the perspective of user education, Deepseek has taken over the baton from Doubao, making a significant contribution to the popularization of the large model industry.
When large language models like GPT-01, Deepseek, and Tongyi Qianwen swept across various industries with their astonishing performance, we once thought that the future of AI was already within reach.
However, behind the “technological prosperity,” there seems to be a deep-seated confusion and contradiction—many seemingly dazzling AI applications ultimately become “chicken ribs”.
Large models appear to have rich functionalities on the front-end interface, but achieving deep integration on the back-end (process management, business system integration, data governance, compliance risk control) often encounters numerous engineering challenges.
Small and medium-sized companies often remain at the “Demo-level integration,” struggling to support continuous and reliable operation in critical business scenarios.
It is reported that, apart from the already collapsed Yingyan Wisdom Chinese Medicine, Forward (an American AI medical company that has received over $657 million in funding from institutions like SoftBank), Afiniti (once an American AI unicorn that completed six rounds of financing)…; the halted Zhujian Intelligence, the acquired Waveform Intelligence; there are many small AI application startups that are on the verge of bankruptcy.
Has the explosion of AI large models truly arrived, or have we yet to touch the real value heights?
Deep Contradictions Beneath Surface Prosperity
In the AI boom, many teams first “gather” technology demos and then look for scenarios. For example, when you see a tool that automatically generates PPTs, do you feel a sense of amazement?
However, when you actually delve into using it, you find that it does not address your real pain points in the creative process—you still need to invest a lot of time in conceptualizing content, adjusting structure, and refining ideas.
Ultimately, the generated PPT may still require multiple adjustments for visualization.
Such solutions do not create a strong dependency; in fact, you may very well view them as “varied but shallow” burdens?
Many applications target “technologically feasible” rather than “truly needed by users” scenarios.
It is precisely the “misalignment” or “pseudo-demand proliferation” in AI technology applications—although large models indeed possess powerful capabilities, the actual effects of their applications often show a significant deviation from the fundamental needs of users.
AI is undergoing a technological leap, but many current products still fail to fully address user pain points, leading to their true value not being completely realized.
In specialized fields, once users discover that the high cost of errors in AI outputs leads to greater caution or even rejection of use.
Minor defects or small mistakes can be tolerated in demonstrations, but in real business environments, they can have serious consequences, leading users to revert to traditional tools or only try AI in “low-risk” scenarios.
Furthermore, even in workflow areas where we believe AI tools are easiest to implement, changing user habits remains a significant challenge.
If it only offers a 10% improvement in efficiency and does not address the most burdensome and complex parts of our work, to be honest, it is difficult to overcome the user migration threshold when faced with tools like Microsoft and CRM that have been used for decades.
The Challenge of Value Capture
Looking closely at the commercialization path of AI, whether it is Microsoft’s Copilot or Adobe’s Firefly, these giants have quickly embedded AI capabilities into our daily tools through bundled sales.
For ordinary users, the free or low-cost services provided by these “big factory ecosystems” have quickly broken the payment model for independent AI applications, greatly compressing the survival space for small and medium-sized companies in the market.
More importantly, the “core value” of AI is difficult to define; most AI applications still exist on the periphery of the toolchain, and the core value that can truly capture user minds remains elusive.
Imagine, although intelligent customer service can easily handle 80% of common problems, the remaining 20% still require human intervention.
This design leads to two major problems: first, a gap between user expectations and actual experiences arises; second, the “replaceability” of AI is still unachievable in many business segments.
This limitation means that while AI has brought initial efficiency improvements, the overall perception of value has not qualitatively changed, even increasingly frustrating consumers.
Intelligent customer service’s “read receipt chaos” and “always unreachable human agents” often drive modern people crazy, perhaps reflecting the most real scenario.
Once users feel that AI cannot genuinely save them significant time, reduce costs, or enhance decision quality, they will choose to return to traditional methods due to “incomplete effects,” further estranging trust between users and AI.
Due to the limited accumulation of industry data among most small and medium-sized companies, they often find themselves in a vicious cycle of “few users → little data → poor performance → user loss,” making it challenging for product optimization and iteration to meet expectations.
Industry Deepening: The Key to Technological Breakthroughs
If AI is to transition from “showcasing technology” to a genuine “breakthrough in value,” it must go beyond simple toolization and delve into every niche scenario of the industry.
In this process, deepening vertical fields and integrating expertise will become decisive factors. In other words, companies with know-how matched with large models will thrive.
For instance, in technology-intensive industries like semiconductor design and new drug development, AI can combine with physical simulation, data models, and specialized knowledge bases to create true industry transformation.
These fields share a common trait—they have very high knowledge systems and data barriers. Only by deeply integrating industry-specialized AI applications can they stand out in genuine technological innovation.
The semiconductor industry optimizes the wafer manufacturing process and improves yield through AI; new drug development relies on AI for molecular screening and prediction. These applications not only enhance efficiency but may fundamentally change the operational models of the industry.
Such breakthroughs in “deep application” are precisely the direction the current AI industry urgently needs.
The Future of Human-Machine Collaboration
The application of large AI models is increasingly moving away from merely “replacing humans” towards enhancing human capabilities.
The value of AI, especially in daily workflows, often manifests as collaboration rather than replacement.
Among them, the rise of Notion AI in the US is a typical example.
Initially, it might only provide convenience when organizing meeting minutes, but as usage scenarios continue to expand, Notion AI gradually evolves into the core of workflow, providing multi-faceted support for content creation, project management, chart creation, and knowledge sharing.
This gradual penetration makes the boundaries between humans and AI increasingly blurred; AI transforms from a “tool” into an “assistant,” even becoming an indispensable part of every decision-making process.
As technology continues to evolve, AI will no longer merely be an external tool but will become an intelligent partner in our work and lives, helping us enhance creativity, optimize decisions, and even provide real-time assistance in complex scenarios. Thus, the concept of intelligent agents can truly take root.
Hardware + AI: New Opportunities for Real-World Applications
If we broaden our perspective, we will find that breakthroughs in AI by 2025 will not be limited to innovations purely at the software level; the heat of AI hardware also proves the importance of application carriers for AI.
With the combination of AR, VR, and smart hardware, the application of large AI models will enter a whole new stage.
Through innovations in hardware carriers, AI can overcome the limitations of traditional desktop applications, expanding its application scenarios from “flat screens” to “holographic displays” and “real-time interactions.”
In addition to consumer products, the deep integration of hardware and AI will also give rise to entirely new industry application models and solutions, opening up more diverse and disruptive application fields for AI.
For example, in the future, maintenance engineers using smart AR glasses combined with AI could receive real-time fault information from equipment and instantly generate troubleshooting plans.
Through voice recognition and visual enhancement technologies, AI can directly provide fault diagnosis guidance to engineers in the maintenance environment. This interactive mode will greatly enhance work efficiency and reduce human errors.
Breaking the Deadlock: Distilling Truth and Integrating Industries
The true way to break the deadlock in AI applications is to shift from “dazzling technological displays” to genuine scene innovation and deep industry engagement.
The commercialization path of large AI models is long and arduous.
It requires breaking through technological boundaries and deeply connecting with industry demands to solve real pain points.
As for the future of AI applications, perhaps it lies not in the widespread adoption of “large models,” but in how to find the intersection of technology and market demand in specialized fields, thereby achieving a true breakthrough in value.
Truly valuable opportunities and paths will be reserved for teams and projects that understand the industry, can access good data, and are capable of pragmatically solving real problems.
From “showcasing technology” to “realizing value,” the path to breaking the deadlock is already clear.
-
Working Parents, Doubao Raising Kids
-
Review of AI Large Models’ “Explosion” During the Spring Festival, Alibaba and DeepSeek at the Table
-
Can Small Models Counterattack Deepseek and OpenAI?
-
AI Search: The “Double-Sided Chess Game” of Technology and Business, Will It Overthrow Xiaohongshu?
-
Revealing: The Story of Musk and OpenAI’s Grievances!