Behind any artificial intelligence, even the most advanced, stands the human mind. The AI Journey conference is like a model of our society in miniature, where you can find everyone who encounters AI in their work in one form or another.
Every year, the event attracts the attention of developers and researchers from around the world. And the 2024 conference was no exception, offering presentations on innovative solutions and tools which will be valuable for IT specialists. AI Journey is a convenient platform where you can learn about all the global trends in AI and IT at once. It also offers industry professionals the opportunity to discuss the hottest issues and biggest challenges facing the international community and the industry.
Developers play a key role in the creation and spread of artificial intelligence. Developers from Russia now have access to GigaView, a new tool based on the GigaCode AI assistant, which boosts the efficiency of teamwork on software projects. The new solution was presented by Andrey Belevtsev, senior vice president and head of technology development at Sberbank. “In today’s world it’s extremely important to provide our developers with effective tools for boosting growth in productivity, and to give businesses the opportunity to efficiently apply these solutions for process optimization. GigaView is aimed at automating the most labor-intensive and routine operations and improving management of the development process. The tool reduces coding errors and makes creating code transparent for all team members”, Belevtsev explained.
GigaView can be used to identify potential defects at the development stage and to assess the quality and overall productivity of development work on AI products. Interestingly, the solution can be used to track the productivity both of the team as a whole and of each team member individually, on the basis of metrics. The tool has been made available on GitVerse, a Russian platform for working with source code. It already supports the Java, JavaScript and TypeScript languages, and more will be added in the near future.
Belevtsev also spoke about improvements to existing solutions. For example, a functional upgrade of Sber’s GigalDE Desktop is due to be released soon. Users of this Russian integrated environment will now be able to solve Java project code tasks more quickly, assisted by AI agents which will greatly reduce time spent on software development. It will be become much simpler for IT experts to find the project data they need. GigaIDE Desktop has already proved its safety and efficiency during testing by Russian government departments, and has been included in a special register.
It may sound like these new tools are only available to people with specialized training, but that is not quite correct. The AI Journey conference showcased solutions for specialists in all sorts of fields, including the creative industries and the world of art. Kandinsky 4.0 Video, for example, is a neural network offering nonstandard creative solutions. During the conference, Sber presented the beta version of the new generation, which can be used to create realistic video clips from a text prompt or starting frame. The AI product is convenient to use and is suitable both for ordinary users generating amateur video clips and for designers, videographers and film animators. Kandinsky can produce a professional clip or project trailer.
Although it’s only a year since Kandinsky was launched – it was first presented at the previous AI Journey conference – its developers have already greatly improved the quality of the video clips it creates. And the videos can now be generated much more quickly. The model can turn a user’s idea into a 12-second video clip in HD resolution of 1280×720. There is also a version for the most impatient artists, known as Kandinsky 4.0 Video Flash, which can produce a video clip in just 15 seconds. This version of the neural network will be available to the general public as early as 2025.
According to Belevtsev, “these models, such as Kandinsky Video, are facilitating global development in ths important field, taking modern technologies much closer to the synergetic level of information processing, perception and creation possessed by human beings”.
The fundamental position of the developers is that many of the technologies in the AI sphere must remain open-source as this, in their view, will greatly accelerate the development of the AI community. An example of this approach is the active publication in open source of Sber’s artificial intelligence models. The codes and weights are now available to all users, particularly the basic and instructive versions of the fastest model, GigaChat Lite, but also GigaEmbeddings-instruct, the best Russian-language model, according to the ruMTEB benchmark.
One of the most important events during the conference was the presentation of an advanced algorithm for accelerating transport routing. Sber and its specialists from the AI Laboratory, in partnership with their colleagues from ITMO University, have created the RIDE Python library, which can be used to build optimal routes much more quickly. What makes the solution particularly noteworthy is that it enhances the performance of logistical calculations from the city level all the way up to the country level. The library is based on an innovative approach to dividing the graph into clusters, thereby affording developers greater control over the quality of their work. The method has been tested on over 500 cities.
As far as rational use of resources is concerned, we know that purchasing goods in online marketplaces can be very time-consuming. But developers have now found a way of helping people to navigate around a huge number of goods and select the best bargains quickly. During the AI Journey conference, new recommendation systems devised by researchers from Sberbank’s AI Laboratory were presented. The researchers analyzed open data sets for training neural networks that help users find products, thereby developing a new approach to checking the accuracy of recommendation algorithms. As a result, if the algorithm is chosen correctly, marketplace users and online services will see more accurate recommendations and save some time – that most precious of resources.
Herbert is a nom de plume.
I am an experienced telecom professional blogging about the fastest growing Internet market in the world: Myanmar.