Sber’s GigaChat neural network model and other proprietary AI-driven solutions will be helping contestants execute tasks
Sber has opened registration for AI Journey Contest, an international online artificial intelligence contest. This year, the prize fund of the event held by Sber and the AIRI Institute will be RUB 8.5 mn (over $90.000).
Andrey Belevtsev, senior vice president, head of Technological Development, Sberbank: “Each year, our contestants from around the world offer cutting-edge AI-driven solutions that expand people’s capabilities, make our lives better, and streamline business processes in companies of different sizes. This time, we have prepared tasks related to large language model development, focusing on multi-modality and the use of advanced AI methods in science and robotics. These are high-potential areas that we promote at Sber and that will shape the future of artificial intelligence. We welcome all challengers and expect the event to yield many disruptive ideas, including those that will help enhance our AI solutions like GigaChat and Kandinsky.”
At AIJ Contest 2024, participants will have to choose one or several of four challenges. In an additional nomination to each of them, the solution inference time will be evaluated for the top 10 teams, which is important in terms of ESG agenda and improving the user experience in working with AI models.
Emotional FusionBrain 4.0
Participants will try to create multimodal models to work with video, audio, and text. The model’s responses across different videos will show how successful it is at “watching” and understanding content. The best approaches can be used to upgrade Kandinsky and create datasets for training generative models.
As an additional nomination to the task, the best multimodal models will be tested on intellectual games video tournaments. The best solutions must not only answer questions about the game and detect the roles of its participants but also figure out who of them is lying.
Multiagent AI
The goal of this task is to demonstrate how artificial intelligence can be useful for scientific research. Participants will have to create a multi-agent RL (reinforcement learning) system whose agents can solve problems by joining together in different cooperation schemes. Building multi-agent artificial intelligence is a SOTA (state-of-the-art) solution in fundamental AI-driven model development. These solutions can be used for modeling complex systems and automating experiments.
Embodied AI
The third task was prepared in cooperation with Sber Robotics Center and involves building a control system for an embodied agent (a robot assistant) based on language instructions and visual information.
Embodied AI’s goal is to create robots that can solve complex tasks that require interaction with the environment and the user, as well as communication with them in a natural language. The task seeks to drive one of the trending areas in this field: creating AI systems that perform complex tasks set by humans with natural language commands. The task is set in the Robotics Center, which will allow participants to try their hand at mining real data and contribute to the development of a promising AI trend.
E-com AI Assistant
This task is aimed at developing an AI assistant based on the GigaChat large language model. This assistant should be able to identify the user’s needs during a dialog with the user and further recommend the most relevant products for purchase on Megamarket. Participants’ solutions will be evaluated using a dialog simulator. The best of the proposed methods can form the basis for the development of recommendation services and specialized AI agents capable of solving the tasks of a particular company or business area, from retail to large industrial enterprises.
18+ participants can apply for AIJ Contest 2024 to solve tasks posted on DS Works until October 28, 2024. The award ceremony will take place at AI Journey, Sber’s international conference on artificial intelligence and machine learning.
Press Release received on Mail
