When Yu Kaichen first began discussing world models in 2023, the reaction was that this field was too far removed, bordering on fantasy. However, three years later, amid the activity of figures such as Fei-Fei Li and Yann LeCun, and the attraction of billions of new participants, world models have become a central theme in the field of physical artificial intelligence.
Yu, who by then was already a world champion in the ECCV autonomous driving track and ranked first in the Fail2Drive benchmark, made a decision that seems least intuitive for a researcher: he abandoned the very methodology that earned him those titles.
His career began at Alibaba's DAMO Academy, where he joined as an Alibaba Star after earning a Ph.D. in AutoML. His work, BEVFusion, which merges LiDAR and camera data into a single physical space, became one of the most cited works in multi-sensor perception and a notable achievement at NeurIPS 2022.
After working at DAMO, he moved to Westlake University in 2023 to establish AutoLab, where he focused on developing generative world models for autonomous driving. Work with a leading Chinese automaker helped dispel old assumptions. Yu concluded that if a model is capable of generating realistic data, it already partially understands the world, and further pixel rendering is merely an expensive detour.
By late 2024, the team shifted direction to an implicit approach. The model stopped rendering pixels; instead, it began constructing representations of physical concepts in an abstract mathematical latent space, decomposing scenes into a finite set of constituent concepts, and combining them to process situations never fully encountered in the training data. Yu compares this to LeCun's implicit JEPA approach, but with a different answer to the question of latent space content: here, composite physical concepts are learned end-to-end.
Internal experiments conducted in 2025 demonstrated a significant increase in success rate on complex tasks compared to baseline imitation and reinforcement learning models, while maintaining a lower prediction cost.
In January 2026, Yu transitioned his research from the lab to the company Awomo, based in Hangzhou. Initially founded only by Yu and Chief Scientist Tun, a group that Yu describes as 'consistently self-verifying geniuses' joined five months later. Awomo successfully closed a seed and angel funding round exceeding 100 million yuan, receiving support from InnoFund, Dongfang Jiafu Fund, Zhengxuan Investment, Tianqi Capital, Westlake Innovation Fund, and Jinma Investment.
Awomo's core premise is that the scaling limit Yu previously helped create does not apply to physical intelligence. A VLA-type model fine-tuned on more demonstrations cannot suddenly grasp a new factory layout. Instead, a model that has learned to decompose a scene into concepts can recompose them on the fly: for example, catching a falling cup while simultaneously closing a door, rather than mechanically completing one subtask before starting the next. The first version, Awomo-v0.1, has already been tested on the RoboTwin 2.0 test track.
Taxpayers can now receive answers to complex tax questions from artificial intelligence, eliminating the need for them to study numerous sources of information independently.
The 'AI SOLIQ' platform, which operates on artificial intelligence, provides entrepreneurs, accountants, and enterprise managers with the ability to quickly obtain simple and practical clarifications on tax issues.
During a press conference held on the 'Uzbekistan' television channel with the participation of the leadership and specialists of the Tax Commission of the Republic of Uzbekistan, the topic 'AI SOLIQ — a new generation digital platform based on artificial intelligence' was discussed. The possibilities of the 'AI SOLIQ' portal, created for taxpayers, the application of artificial intelligence technologies in the activities of tax authorities, the legal aspects of using these technologies, as well as providing quick recommendations for lawful business activities were reviewed.
The ai.soliq.uz portal was launched by the Tax Commission. It unites new generations of tax services created using artificial intelligence technologies on a single digital platform. Initially, the platform is intended for entrepreneurs, accountants, and company managers. Users can conduct an intellectual analysis of their organization's activities in a short time, identify the reasons why electronic invoices are categorized as 'red', and receive necessary explanations regarding tax rating, tax benefits, and tax reporting.
The services on the platform function in a dialogue format, that is, a chat. This allows for receiving information on complex tax issues in a simple, understandable, and practice-oriented manner. The importance of 'AI SOLIQ' is not limited only to providing services to taxpayers. The portal also contains the main works of the Tax Commission in the field of artificial intelligence. In addition to implemented digital solutions, it presents the technological foundations being developed for the digital transformation of the tax administration. These include solutions based on the national language model 'Soliq LLM', speech and text generation technologies, computer vision, and machine learning for big data processing.
These technologies play an important role in processing and analyzing data in the tax sphere, automating routine processes that require human labor, and improving decision-making processes. Today, artificial intelligence technologies are used not only in the services provided to entrepreneurs and taxpayers but also in all links of the tax authorities' activities.
In the system, ranging from local employees to the leadership of the Tax Commission, methods of using artificial intelligence for data analysis, automation of repetitive processes, improvement of the quality of management decisions, and increase in employee efficiency are being implemented. This approach contributes to the gradual elimination of processes in the tax administration that are overly dependent on the human factor, and the development of a data-driven management system.
The creation and development of the artificial intelligence ecosystem in the tax sphere is also related to the activities of the Artificial Intelligence Laboratory, established at the initiative of the President at the beginning of the current year. This laboratory functions as a single center responsible for developing, implementing, and developing AI-based solutions aimed at improving the tax administration and creating modern digital services for citizens and business representatives. The solutions developed by the laboratory are aimed at both digitizing existing processes in tax authorities and improving the quality of taxpayer service.
The wide implementation of artificial intelligence in the tax sphere should primarily contribute to simplifying the interaction between citizens and entrepreneurs with tax authorities. The easier it is for an entrepreneur to obtain necessary information, understand tax reports, learn about the conditions for using benefits, or identify the reason for a specific problem, the more opportunities they have to voluntarily and legally fulfill their tax obligations. From this point of view, 'AI SOLIQ' should be seen not just as a digital service, but as a technological solution that moves the interaction between the taxpayer and the state to a new level.
The Tax Commission plans to expand the use of modern artificial intelligence technologies and increase the range of intelligent services for taxpayers in the future. The main goal of this process is to reduce excessive time and bureaucratic obstacles in tax matters for businesses and citizens, make interaction with tax authorities simpler and more convenient, and bring the quality of tax services to a new level.
