Days ago, Beijing announced two technological surprises that caused a stir in Silicon Valley. The first was the birth of a giant artificial intelligence language model called "Kimi Ki 3" from a startup company, which underwent the world's most important programming tests and outperformed some models from "Anthropic" and "OpenAI," proving its exceptional ability to analyze massive documents, reports, and long databases more than usual, which is attractive to companies.
The second surprise has a deeper impact due to its uniqueness. It is the launch of the "QiuShi" engine, developed by a team from "Zhejiang" University, as a specialized agent for independently conducting scientific research. It also outperformed "Open Science Desktop" and "Claude Code" systems from the American company "Anthropic" in the "Research Clo Bench" test, which was specifically designed to measure this type of intelligent skill and involved international companies.
For the first time, artificial intelligence proved its ability to act as a researcher who plans, conducts experiments, and corrects his own mistakes, like a graduate student. Therefore, the system was described as a "milestone" after successfully conducting a study by performing thousands of steps, from hypothesis formulation to experimental verification.
What concerns Silicon Valley giants is that China is no longer an imitator but is opening new fronts, with its comprehensive technological strategy. It is redefining standards and bringing about profound transformations by introducing a new way of building AI and its real-world applications.
The unveiling of the "Kimi Ki 3" system caused a real shock. It has become one of the largest open models, with a size of up to 2.8 trillion parameters, capable of handling very large and unprecedented contexts.
But more important than its skill in performing complex tasks and its high efficiency in utilizing resources is its low price, which is reminiscent of the cost-effective "DeepSeek" that shocked everyone on its announcement day. Therefore, even if Chinese models are not the smartest ever, their economic cost and open-source nature pose a real challenge and an attractive proposition for companies using AI.
The achievement of "QiuShi" attracted attention because it does not answer a scientific question based on what it has read or been provided with, but rather performs a series of interconnected steps, conducting experiments itself, analyzing results, and presenting conclusions and suggestions that were not previously put forward.
Chinese developers did not claim that their model would reliably succeed in all future tests it would undergo, and they admitted that it still needs training, qualification, and human intervention. However, it is an impressive step in scientific research and represents the beginning of a shift from "AI that assists the researcher" to "AI that leads the research itself," which is significant as it will make research advance at an unimaginable speed.
The two models reveal one aspect of China's multifaceted technological plan, which attempts to cover various fields and is not content with excelling in building language models. "Kimi Ki 3" competes in a known arena, while "QiuShi" is a breakthrough in a different arena, making us wonder: Can a machine become an actual researcher, or will it remain merely an assistive tool? And the success of both projects in close proximity by two completely different teams in their nature of work, i.e., through a startup on one hand and an academic university team on the other, is what gives the topic its importance and shows us that we are not facing two isolated announcements, but rather this is an indicator of an integrated system moving in multiple directions.
China has trained some of these advanced models on local processors from "Huawei," "Alibaba," and others. It still bets that through superior engineering skill, it can compensate for its lack of computing power and its weakness in the field of chips.
In China, nothing happens by chance. "QiuShi" was launched simultaneously with the World Artificial Intelligence Conference in Shanghai, where Chinese President Xi Jinping announced the start of a new global alliance to regulate AI to serve "common prosperity," with the clear goal of portraying China as a global technological power, not just a marginal competitor.
The next phase will be exciting because the competition will not be limited to developing smarter language models like "ChatGPT" and "Claude," as that has become self-evident. The real race will be to launch new super-intelligent specialized agents in science, medicine, and engineering, similar to "QiuShi." And perhaps another issue will come into play: who is more worthy of users' trust, in terms of safety, reliability, and personal privacy.
From "Asharq Al-Awsat"
شارك برأيك
Two Surprises from China