The Paris AI Action Summit revealed a stark shift in who holds power in the creative stack: as models shrink and open-source tools democratize access, the gatekeepers are losing their grip on compute.

Paris, Feb. 12 (Xinhua) — Topic: Paris Summit Offers a Pulse on the New Trends in Artificial Intelligence Development

By Luo Yu of Xinhua News Agency

From February 10 to 11, the AI Action Summit was held in Paris. Inside the Grand Palais, the main venue, political and business leaders from multiple countries gathered for heated discussions on global AI governance. In the “Station F” startup park in southeastern Paris, thousands of AI entrepreneurs and innovators were busy showcasing the latest applications of AI and negotiating commercial partnerships. At the Shangri-La Hotel along the Seine River, experts and scholars from China, the United States, Canada, and other countries engaged in in-depth exchanges regarding AI technologies and their applications.

This event was not merely a display of cutting-edge technological exploration but also a profound outlook on future development trends. Currently, the depth and breadth of AI technology are continuously expanding, evolving toward richer, more efficient, and application-oriented directions. Particularly at the technical core level, the “slimming revolution” of large models is breaking through computing power constraints. As one expert put it, these models are undergoing a transformation from “clumsy athletes” to “agile dancers.”

“The future development of AI will trend toward terminalization and lightweighting. Large models need further miniaturization and lower energy consumption to be deployable on mobile phones and wearable devices, with personal agents and digital assistants serving as typical application scenarios,” said Liang Zheng, professor at the School of Public Policy and Management at Tsinghua University and deputy dean of the Institute for AI International Governance at Tsinghua. “With advancements in technologies such as multimodal learning and reinforcement learning, the large-scale deployment of service robots, autonomous vehicles, and drones will become possible.”

I think lightweight models on personal devices could bypass corporate platforms, giving creators direct control over their digital assistants.

AI hardware and software innovations are also advancing at a stunning pace. Xu Wei, professor and deputy dean of the Interdisciplinary Institute for Information Science at Tsinghua University, believes that short-term visible trends include “expanding existing large language models into multimodal capabilities” and “integrating AI into various workflows.” On the other hand, the scientific research community is exploring new model architectures that surpass current large language models. The core of this lies in richer modeling capabilities, providing AI with a deeper understanding and expression of the world.

For creators, integrating AI directly into existing workflows reduces friction but raises questions about data privacy for independent creators.

This summit placed particular emphasis on promoting open access to AI resources and lowering the threshold for developing countries to acquire AI technology. In this regard, China’s large language models have stood out. Represented by DeepSeek, they exhibit characteristics of openness, low energy consumption, and transparency, becoming a new paradigm for inclusive global benefits. The French newspaper Le Monde reported that DeepSeek’s debut was stunning; relying on less computing power and data consumption, as well as lower training costs, it achieved performance comparable to the latest products from the United States. This achievement is of extremely significant importance in promoting AI development in countries other than China and the U.S., and in advancing the inclusivity and broad accessibility of AI applications.

On licensing, open-source models with lower training costs allow smaller studios to compete without relying on expensive enterprise licenses.

Outside and inside the conference venues, the open-source mindset of Chinese enterprises received high praise. Maxim Carrel, a data scientist at Scalian, a French digital systems company whose firm is dedicated to promoting small-scale, specialized open-source AI, stated: “We are chasing the best AI. For example, we are testing Alibaba’s Tongyi Qianwen model; it is excellent—both efficient and low in energy consumption.” Dan Carpenter, sales vice president of a British AI cloud service enterprise, described DeepSeek as a “surprise” and expressed hope that “the more open-source companies like this exist, the better.”

In the current era of rapid AI development, voices from the Global South are often overlooked. At the summit, multiple experts repeatedly emphasized the inclusivity of AI governance, calling for greater representation of the Global South and other groups in AI governance dialogues. Sean O’Hegarty, a senior researcher at Cambridge University, stated: “Many Chinese scholars emphasize the importance of using technology to help achieve sustainabl

Paris Summit Highlights the New Pulse of AI Development

The stakes at the Paris AI Action Summit were clear: who gets to shape the rules? While Western powers signed declarations on safety, the Global South argued for representation in governance. This isn’t just diplomacy; it’s about whose data trains the models and whose values dictate their use. If we ignore developmental needs from the South, we risk building an exclusive tech stack that serves only a few.

The Push for Inclusive Governance

A recurring theme was the urgent need to include experts from the Global South in AI governance dialogues. As one perspective highlighted during the summit: “AI development should fully consider the developmental needs of Southern countries, involve their experts, understand their requirements, and empower them with a voice in global decision-making.”

I think ignoring Global South voices creates blind spots in model training data that affect local creators. For creators, centralized governance often overlooks the specific licensing needs of non-Western artists.

Defining “AI for Good”

The concept of “AI for Good” dominated the conversation, with a prevailing view that development must align with fundamental human interests and value standards. This requires joint promotion of development, joint protection of security, and shared benefits.

Representatives from France, China, India, and the European Union jointly signed a declaration vowing to ensure artificial intelligence is open, inclusive, transparent, ethical, safe, reliable, and trustworthy. I read this as a positive response to calls for ethical AI, though implementation remains the true test.

On licensing, vague promises of “transparency” do little to protect creators from unauthorized scraping. I think without clear commercial use policies in these declarations, creator revenue streams remain uncertain.

The Path Forward

The summit showcased diverse innovation and open development trends, but also highlighted the urgent need for global collaboration and an inclusive governance framework. Facing infinite opportunities and numerous challenges, I followed the sentiment that the international community must join hands to deepen innovative cooperation.

We need a system where AI truly becomes a great force benefiting all humanity, rather than just reinforcing existing power structures. The Paris summit was a start, but the real work lies in ensuring these governance systems are accessible and equitable for every creator worldwide.