Source: Tencent Technology
Author: Zhang Lu

During the Chinese Lunar New Year, DeepSeek's explosive popularity spread to Silicon Valley, causing anxiety among many technology giants in Silicon Valley. OpenAI CEO Sam Altman admitted that OpenAI's closed-source strategy has placed them on the wrong side of history, prompting a reevaluation of OpenAI's open-source strategy. Zuckerberg also acknowledged at the company's all-hands meeting Q&A that DeepSeek has achieved a novel breakthrough.

The Chinese original large model DeepSeek has also convinced everyone that AI will accelerate changes in the world.

In fact, over the past few years, AI technology has been rapidly advancing, continually surprising the world.

Standing at the beginning of 2025, it is a pleasure to share my thoughts with everyone. This year, I expect five major AI trends to gradually take shape:

First, small models in vertical fields will quickly land in the industry.

Second, more technological innovations will emerge at the infrastructure level of AI, especially in terms of reducing energy consumption and costs.

Third, more small models are beginning to enter edge devices.

Fourth, the open-source ecosystem will continue to thrive, with open-source models becoming increasingly rich and diverse.

Fifth, new Algorithm Models and architectures are constantly emerging.

At the same time, the widespread application of AI also makes us continuously examine our relationship with AI. AI makes our work more 'intense' and also makes many tasks easier. As AI tools become more popular, competition has intensified, but in some ways, it has liberated our creativity, especially in the fields of art and productivity.

AI is gradually replacing some traditional jobs. However, the more critical point is that those who can skillfully use AI tools will replace those who cannot effectively use AI tools.

In the face of the rapid development of AI, we also need to acquire new skills and knowledge. In addition to learning how to use AI tools, we need to master key abilities such as questioning and problem disaggregation. The most indispensable ability in the AI era remains decision-making capability.

The five major AI trends forecast.

• Forecast one: Vertical small models will quickly land in the Industry.

This has actually begun to happen in Silicon Valley and the industry.

In the past, everyone discussed large models, and as the scale of the models continued to expand, the training effects were improved by accessing a large amount of data and parameters. However, over the past year, the focus has gradually shifted from the rapid growth of AI models to practical applications - that is, how AI technology can be quickly implemented and achieve large-scale industrial applications.

In this process, the large-scale application of small AI models will definitely become a trend.

For many specific industries or vertical fields of AI applications, although the data volume needs to reach a certain scale, the quality of the data is even more critical. In fact, the quality of the data may be more important than the quantity. By using high-quality industry data to train small models in vertical fields, not only can the accuracy and effectiveness of AI applications in specific scenarios be improved, but more importantly, small models are smaller in scale, lower in cost, more energy-efficient, and require less from GPUs.

These characteristics make small models more in line with the industry's demand for AI technology applications in terms of cost. Especially in scenarios aimed at enterprise customers (B-end applications), technology costs are often the most critical factor. In this context, the 'miniaturization' of models becomes particularly important.

• Prediction two: More technological innovations in the AI infrastructure layer will emerge, particularly in terms of reducing energy consumption and costs.

The field of artificial intelligence will see more technological innovations at the infrastructure level, especially focused on how to reduce energy consumption and costs. In fact, many related technologies have already begun to be implemented.

We know that one of the biggest challenges facing AI development is high costs, high energy consumption, and over-reliance on GPUs, which prevents meeting the demands of industrial applications. Based on this, it will drive the next wave of more practical and commercially viable AI applications.

Currently, many AI infrastructure technologies can help optimize these applications, thereby significantly reducing GPU consumption on a large scale. This optimization not only includes the reduction in GPU consumption but also encompasses decreased energy consumption, fundamentally lowering overall costs.

For example, innovations in some Hardware and Software technologies have already reduced energy consumption by 15 to 100 times, or even more. At the same time, GPU consumption can also drop to about 1/4 of its original level, or close to 1/10. These are all potential trends for future development.

• Prediction Three: More small models start to enter edge devices.

By 2025, an important trend will be the application of AI in edge devices. In fact, this trend has already begun to emerge in Silicon Valley.

The application of AI in edge devices has been driven by numerous large and small enterprises. This will give rise to new AI interfaces, representing entirely new forms of interaction brought about by AI. In the past, our modes of interaction mainly relied on traditional devices like smartphones, but with the proliferation of AI, future modes of interaction will no longer be limited to traditional edge devices like smartphones. More devices will integrate with AI, such as smart glasses, projectors, speakers, lights, and various small everyday devices. These devices will become new carriers of artificial intelligence.

For example, a company we recently invested in—Nexa AI—has AI small models that can run efficiently on edge devices like Raspberry Pi, and its generative AI performance is akin to that of GPT-4. From this perspective, we can see the tremendous potential of this technology in edge devices.

Furthermore, with the widespread adoption of artificial intelligence, new AI interfaces are bound to emerge. For instance, on the consumer side, AI smart glasses are gaining attention in the industry.

On the business side, the application forms of edge-side AI are more diverse. For example, in the Logistics and supply chain industry, smart Sensors can serve as carriers of edge-side AI, widely used in various machinery and equipment; in the burgeoning space Technology field, each Satellite can act as an intelligent agent, serving as a carrier for edge-side AI; in the Medical field, various Smart Healthcare devices, Sensors, and even medical instruments can also carry artificial intelligence, becoming carriers of AI.

Therefore, the carriers and interfaces of AI are not limited to mobile phones and computers; it will be embedded in various industries, especially in the industrial sector, where AI will be integrated into different Sensors and Hardware in the form of intelligent agents. The application scenarios for AI are very extensive.

• Forecast four: The open-source ecosystem will continue to thrive, with open-source models becoming more diverse and rich in form.

We remain Bullish on the development of the open-source ecosystem and firmly support its growth. The open-source ecosystem has made tremendous contributions in the field of AI. In fact, since 2024, various open-source platforms have been dynamically active on a Global scale, not limited to the USA; China also has many active contributors to the open-source ecosystem, such as DeepSeek, which has recently sparked heated discussions in the global tech community. The recently released Llama 3.1 also fully demonstrates that the open-source ecosystem can promote the generation of more language models and smaller models.

Looking ahead to 2025, it is believed that the open-source ecosystem will continue to thrive and become an important platform for supporting and incubating startups. Many technologies in vertical fields have originated from open-source, further proving the key role of the open-source ecosystem in promoting technological progress and innovation.

• Forecast five: New Algorithm Models and architectures are continuously emerging.

Currently, discussions about generative AI and large language models are very active, but at the same time, many new Algorithm Models and architectures are also continuously emerging. For example, $Alphabet-C (GOOG.US)$ and$Microsoft (MSFT.US)$Recent research papers are exploring new Algorithm Models and architectures. A significant feature of these new models is that they can operate efficiently not only on GPUs, but some models can even perform better on CPUs. This discovery may have a profound impact on the market, as it raises an important question: Do all AI applications have to rely on GPUs? Or will some Algorithm Models perform better on CPUs in the future?

These changes and trends are quietly occurring, indicating that they will have a profound impact on the entire Industry.

AI makes people more "competitive" and also more "relaxed".

As AI continues to penetrate every aspect of life, the relationship between people and AI is quietly changing. It can be said that AI makes us both "more competitive" and "more relaxed."

For example, were you more competitive before the computer existed, or after you got one? Clearly, you were more "competitive" after getting a computer. This is because various tools help us improve efficiency and liberate productivity, thereby accelerating the progress of all tasks. With the enhancement of productivity, everything is accelerating, innovation is accelerating, and growth is accelerating. Every Industry is rapidly developing under the empowerment of AI. And acceleration means competition is becoming more intense. Therefore, AI will not only promote industrial upgrades but also make us face more intense competition on an individual level.

I would also like to share an interesting phenomenon. Today's college students, especially freshmen and sophomores, and even some high school students, spend a significant amount of time using AI tools every day. They spend about 70% to 80% of their time using AI applications on their phones. For example, many students hardly use traditional Search Engines like Google anymore, but turn to platforms like ChatGPT and You.com for their searches.

This change is quite interesting. Because we grew up in a mobile-based environment, for us, touch screens, searches, and applications are all natural and do not require special learning. For them, AI applications have already become second nature, similar to how we used touch screens and Search Engines back in the day.

Additionally, it has been found that the younger generation is more willing to use AI tools and interact with AI. At the recent JPMorgan Healthcare Conference, I spoke with many leaders from large companies in the medical field who shared an interesting phenomenon: Many companies have begun to use AI-driven mental health tools, particularly applications aimed at alleviating anxiety and maintaining mental wellness. These companies offered patients two options: one is to interact with AI, receiving feedback and suggestions from it; the other is to communicate online with a psychological counselor. The results showed that 70% of people preferred to choose AI as their mental health advisor, willing to share sensitive personal information and psychological issues with AI rather than a real person, which is astonishing.

This also indicates that the relationship between humans and AI might change faster than we imagine. From initial misunderstanding and resistance to gradual cooperation and now dependence, in the future, AI may become an integral part of daily life, much like smartphones, and new habits will emerge.

What skills and cognitive abilities do humans need in the era of AI?

So, what skills and cognitive abilities do we need in the era of AI?

First and foremost, it is crucial to realize that AI will replace some traditional jobs, but more importantly, those who can proficiently use AI tools will replace those who cannot effectively use them.

With the widespread use of AI technology, new job opportunities will constantly emerge, while many traditional positions may be replaced. In this process, the rapid development of technology will play a decisive role. Just like how we learned to use computers in our childhood, today, mastering the use of AI tools has become an essential skill. In the past, knowing how to use a computer was a basic requirement for job hunting; today, mastering how to use AI tools will also become a foundational capability.

(1) The ability to ask questions and break down problems is crucial.

In the era of AI, possessing the ability to ask effective questions is particularly important. The ability to ask questions can sometimes be more critical than the ability to answer them. By posing clear and targeted questions, AI can provide more precise and valuable feedback and support.

Another crucial ability is to break down complex problems into smaller tasks or to decompose a large work structure into more manageable subtasks. This is not just a skill; it embodies leadership and management capabilities. For example, one of the core responsibilities of a manager in a company is to break down complex work into specific small tasks and then assign them to team members for execution. Nowadays, these tasks may be assisted by AI rather than being directly completed by junior engineers. If a person can only execute tasks but lacks the ability to think, ask questions, and plan, this may pose risks.

Therefore, how to think effectively, decompose problems, and collaborate with AI has become a very important ability.

(II) AI Promotes the Release of Artistic Creativity

I have also observed that one significant advantage of AI is its wide application in the fields of art and creativity. Although some people worry that AI will replace artistic creators, from another perspective, AI is helping more people achieve creativity, similar to the invention of the camera. The camera allows us to capture beautiful moments without the need to learn painting skills; similarly, AI is releasing more creativity, enabling everyone to easily engage in artistic creation.

One of the most important abilities of an artist is to possess creative thinking and express that creativity through strong memory and skills. Photographers did not exist before the invention of the camera, but now, mobile phones have become tools that everyone can use to capture beautiful moments, greatly lowering the barriers to creativity. AI is developing in a similar direction, allowing more people to easily create and express themselves.

Today, even if you have not learned to paint, you can present your ideas and creativity through AI. You may have rich emotions that you could not previously express through paintings or songs. Now, with AI tools, you can express your feelings through lyrics, melodies, and other forms. This technological advancement has greatly expanded the boundaries of artistic creation, allowing more people to transform their inner emotions into artistic works without being constrained by traditional creative skills.

(III) Decision-Making Ability: The Core of the AI Era

The impact of artificial intelligence in various fields is evident, whether in music, fine arts, or other industries, similar changes are occurring. Taking the financial industry as an example, in the past, successful investors often relied on obtaining data that others could not access and making accurate market judgments through data analysis. This required strong analytical abilities, such as Excel skills and data analysis capabilities. But today, AI technology is narrowing this gap. Anyone can access more data and conduct in-depth analysis through AI.

However, the most critical and indispensable ability in the AI era remains decision-making capability. Although AI can provide us with a vast amount of information and perform precise analyses, how to make informed decisions based on these analyses is still the core capability. It can be observed that many skills fundamentally reflect decision management ability and leadership. Only on this foundation can AI's assistance truly unleash its maximum potential.

Appendix: Author Introduction

Zhang Lu, founding partner of Fusion Fund, a renowned investor in Silicon Valley and a successful serial entrepreneur, graduated from Stanford University's School of Engineering. In 2015, he established Fusion Fund, currently managing nearly $0.5 billion in capital, focusing on investments in emerging technology startups in the USA, particularly in the fields of digital transformation and Smart Healthcare. He has invested in over 100 companies, with several of the invested companies going public or exiting through mergers and acquisitions.

Editor/rice