Chinese Large Language Models Become Globally Competitive Thanks to DeepSeek Achievements

The team at the Institute of Automation of the Chinese Academy of Sciences (CASIA) has introduced a world model called PhiZero. This model separates the task of understanding physical dynamics from the task of rendering frames. The paper, titled 'PhiZero: A World Model Built Around Physical Language,' was published on arXiv and began circulating among the embodied AI community this week.

World models that sequentially predict future video frames tend to merge two substantially different problems into one prediction space. The appearance of the scene, including its textures, lighting, and background, uses the same parameter capacity as the dynamics that move objects over time. The CASIA group argues that this mixing is why pixel-based world models often generate visually convincing frames that deviate from physical plausibility when viewed consecutively.

PhiZero divides the prediction pipeline into two stages. The first stage takes the current world state and an action, then outputs a short sequence of discrete tokens, which the authors call 'physical language.' The second stage uses these tokens as a compressed motion plan and only then renders the subsequent frames. The tokenizer module, built upon a transition-level Q-Former and Finite Scalar Quantization (FSQ), generates these discrete tokens. The diffusion decoder reuses the appearance of the first frame as high-frequency prior knowledge, allowing the language tokens to focus their capacity on motion rather than texture.

The compression ratio is quite significant. For a four-second clip consisting of 33 frames, PhiZero uses 256 physical language tokens drawn from a vocabulary of about 25,000 units. In comparison, VAE Wan2.2 requires 44,800 continuous visual tokens for the same clip. This represents an approximate 175-fold reduction in the number of tokens that the world model needs to consume and predict. The reconstruction quality remains comparable to other highly compressed video tokenizers.

The implications of this approach are structural. If world models can reason about short, discrete state trajectories rather than long, continuous streams of pixels, robotics, autonomous driving, and embodied AI agents will gain an understanding closer to programming language than to video codec. PhiZero's contribution lies not so much in the figure of 175 times, but in betting on the idea that a physical language, much like natural language, can serve as the foundation upon which world models compose their decisions.

China added a new category of satellites to its orbital fleet at the end of July. Using the Dongfeng rocket, Chenguang-1 was launched from a commercial zone for innovative space tests, carrying an eight-card computing server designed to perform AI inference in orbit. This mission is the country's latest demonstration in space computing, as it transforms satellites from simple cameras into autonomous agents capable of filtering, recognizing, and making decisions before data is transmitted to Earth.

Luo Yaoci, Chairman and CEO of Zhongke Tiansuan, explained this transformation in simple terms. Traditional satellites capture raw images and transmit them to ground stations for processing. Less than five percent of the collected data reaches the ground, and processing useful information can take days. Satellites equipped with their own computing power can extract valuable insights from the images they capture and send only those insights, significantly reducing bandwidth costs.

The economic significance of this trend extends beyond remote sensing. If a global layer of orbital computing becomes a full infrastructure, it will also provide access to computing resources for regions lacking fiber-optic lines and data centers. Zhongke Tiansuan is preparing a fully domestic POPS-class onboard computer, scheduled for delivery by the end of 2026, and has completed the development of an eight-card space server prototype for training based on domestic CPU and GPU architectures. The company's long-term goal is to create a 10,000-card space supercomputer, which they have named the Tiansuan Plan.

China's system involves many participants besides one company. The National Supercomputing Center in Tianjin presented the Tianhe Joint Space Superintelligence Integration Center to the public at the World Intelligence Industry Exhibition in May 2026. GoSharpened Aerospace, specializing in AI satellite internet, supplies its orbital computing group, StarCompute Plan, consisting of 2800 satellites. The Institute of Computing Technology of the Chinese Academy of Sciences has developed a high-performance heat-dissipating computing module sourced domestically, as well as a radiation-hardened RISC-V multi-core processor. Universities in Shenzhen and Shanghai have already incorporated courses on space computing, and the Hangzhou Institute of Advanced Studies at UCAS has opened a specialized research center for space computing.

China is not alone in these developments. SpaceX has launched test Starlink satellites equipped with AI chips, Google has outlined plans for creating space data centers, and other American players are working on their own orbital computing modules. However, a distinguishing feature of China's position is the breadth of the parallel infrastructure assembled and the speed at which engineering experiments are transformed into the named orbital infrastructure.

A small worm, identified as Halicephalobus mephisto, found in a gold mine in South Africa, has changed the scientific perception of the limits of life on the planet. This animal was cataloged in 2011 as the first known nematode to reside in such deep regions of the Earth's crust.

The discovery began in 2008 when researchers analyzed thousands of liters of groundwater extracted from rocks located approximately 1.3 km below the surface. This investigation demonstrated that complex ecosystems can thrive in environments characterized by being dark, hot, and with little interaction with the surface.

The analysis of this 'devil worm' indicated that certain forms of life are capable of surviving without directly depending on sunlight or an oxygen-rich atmosphere. These results brought new perspectives on the adaptability of living beings and on the environmental conditions that support the existence of life.

For many decades, the predominant belief among biologists was that animal life would have difficulty at great depths; until the 1980s, the majority hypothesis suggested that just a few centimeters below the ground constituted a limit for more complex organisms.

This panorama began to change in the 1990s with the emergence of evidence of microorganisms living kilometers below the surface. From the 2000s onward, biologist Gaetan Borgonie, a specialist in nematodes, began studies to verify if worms could also occupy these extreme niches.

This search led Borgonie and other specialists to the Beatrix mine in South Africa. There, water flowing through fissures in the underground rocks was examined. The team used filters to capture microscopic organisms and, after analyzing over 6,000 liters of water, located a novel worm for science.

The research was expanded to other mines in the country. In one of these explorations, scientists filtered over 12 million liters of water, finding various species of nematodes, as well as fungi, invertebrates, and other microorganisms.

The specimen named Halicephalobus mephisto possessed a crucial characteristic for its scientific study: it was a female capable of producing offspring without resorting to sexual reproduction. Before dying, the animal produced eight viable eggs, which allowed researchers to continue experiments in the laboratory.

The descendants were sent for analysis conducted by geneticist John Bracht of the American University in Washington, United States. Under controlled conditions, it was observed that the worm exhibited preferences distinct from those of other known nematodes. While species like Caenorhabditis elegans depend on decaying organic matter and can be studied at lower temperatures, the devil worm shows better performance near 37°C, a temperature similar to its subterranean habitat.

The species also has a unique mode of locomotion. Instead of swimming, it uses body structures to attach itself to surfaces, a characteristic that may aid its persistence in restricted spaces between rocks.

DNA analysis of Halicephalobus mephisto provided clues about its survival mechanisms. In 2019, the team led by John Bracht identified a high concentration of genes linked to the production of heat shock proteins. These molecules act by protecting other proteins in the organism against damage caused by high temperatures, signaling a specific adaptation to life in hot, subterranean environments.

Subsequent research, conducted in 2024, focused on cytochrome c oxidase, a molecule involved in using oxygen for energy production. Tests showed that this structure operates more efficiently at elevated temperatures and decreases its activity in colder climates.

According to John Bracht, a genomic researcher at the American University, this characteristic can be seen as an evolutionary strategy of the animal. He explained the metabolic behavior of the devil worm, stating that 'They are ensuring they will reproduce more in the better, warmer environment.'

The ability to reproduce without a partner is also being studied as a potential advantage in an environment where encounters between individuals of the same species may be uncommon. Bracht assesses that this tactic would ensure the continuity of the species even in isolated locations. However, he also considers the possibility that males of the species have not yet been discovered, which would allow for sexual reproduction in certain circumstances. The lack of genetic diversity resulting from asexual reproduction is often pointed out as a possible evolutionary disadvantage.

The origin of the organisms inhabiting the depths remains a mystery. Studies conducted by Gaetan Borgonie and geobiologist Cara Magnabosco suggest that some nematodes may have reached these areas through the movement of groundwater, possibly influenced by seismic events.

Studies on the so-called deep biosphere indicate that microorganisms below the surface represent a considerable part of the planet's microscopic life. Estimates cited by researchers suggest they may account for a fraction between 12% and 20% of the total terrestrial microbial biomass.

Another example investigated is the bacterium Desulforudis audaxviator, found in the depths of three continents. Researchers noted that this species shares great genetic similarity between distant populations and may have been present in these environments since the separation of the supercontinent Pangaea, which occurred about 165 million years ago.

The study of these subterranean beings also aids in understanding the origins of life on the planet. Scientists assess that deep environments may provide clues about the conditions necessary to sustain organisms before the emergence of current ecosystems.

The research on the devil worm reinforces a broader conclusion: life has the capacity to occupy seemingly improbable spaces and remain active even under conditions very different from those found on the surface.

Van Heerden, an environmental scientist from the South African bioremediation company iWater, stated in an interview with the team responsible for the discovery: 'As human beings, we think we rule the world, but that is not true.' She added that the study provided a perspective of humility before organisms that have existed for millions, and perhaps billions, of years.