Breakthroughs in Generative AIThe leading technology companies and researchers continued to push the boundaries of generative AI capabilities. OpenAI keeps the global community in anticipation of its next giant: a model unofficially named GPT‑5, which many expect to see this year. According to technology press, the new GPT‑5 may turn out to be a “scale leap” compared to the current GPT‑4. OpenAI’s CEO Sam Altman hinted that as the new version is released, the current GPT‑4 will seem weak in comparison. The upcoming model is expected to become truly multi‑modal – capable of simultaneously working with text, voice, images, and even integrating tools like search and code, which effectively brings it closer to the status of an “omnimodel” AI capable of independently searching and analyzing information on the internet. Such an “agent‑style” AI, capable of fact‑checking and reaching out to external data, could significantly improve response accuracy and reduce the number of hallucination errors.
OpenAI is not the only one setting the tone. Meta is also accelerating its advance on the advanced AI front. In July, Mark Zuckerberg announced the creation of a new Superintelligence Lab at the company and brought in Shenzhia Zhao – one of the co‑authors of ChatGPT and GPT‑4. Zhao headed the scientific direction of this Meta lab project, aiming for long‑term research in the field of artificial general intelligence (AGI). Notably, Meta openly declares a commitment to open source: Zuckerberg intends to publish meaningful research results in open access, which brings both praise and concern within the community. Aggressive hiring of talent from competitors and multi‑billion‑ dollar investments (recall: Meta recently acquired a stake in the startup Scale AI for $14.3 billion) indicate an intensified arms race in AI, where leading firms strive to lure the best specialists and not fall behind each other.
Meanwhile China is demonstrating its own breakthroughs. The young Chinese company DeepSeek recently caused a sensation by developing a relatively inexpensive AI model that in quality of work is comparable to the best Western systems. NVIDIA’s CEO Jensen Huang, during a visit to Beijing, noted that models from DeepSeek, as well as giants such as Alibaba and Tencent, are already “world‑class.” At the World AI Conference (WAIC) in Shanghai in July, China showcased about 40 large language models of its own production and a whole spectrum of derived products. It is clear that China is rapidly catching up to the USA: Beijing has declared a goal to become a global leader in AI by 2030, despite US sanctions and export restrictions on advanced chips and equipment. The scale of efforts is impressive – more than 800 companies participated in the Shanghai forum, presenting over 3,000 high‑tech developments, including 60 intelligent robots of various types. Many of these robots – humanoid and service – point to a growing intention to deploy AI in real‑world tasks from industry to healthcare.
It can be said that July confirmed: generative AI is entering a new phase of maturity. On the one hand, models of the next generation with unprecedented capabilities are approaching; on the other hand, the spirit of open experimentation and competition persists. An interesting dynamic is emerging: while European startups such as the French Mistral AI strive to compete with American leaders (Mistral, together with major players, even called for softening the rush for rigid AI regulation in the EU), Chinese companies challenge on efficiency and cost of model development. All of this pushes the industry to new heights – and makes the question of the rules of the game even more relevant.
