I just watched the speech by Nvidia’s CEO at the Davos Forum — the guy talked about three key breakthroughs in AI over the past year, and there really is something to discuss.

First, how AI models have evolved. They used to produce a lot of hallucinations, but over the past year there has been a serious shift. Models have started working on real research tasks, demonstrating the ability to reason, plan, and answer without prior preparation. In essence, this has led to the birth of agentic AI — models that can act independently.

Second, it’s the explosive growth of open models. DeepSeek became the first serious open-source inference model, and it turned out to be a game-changer for the industry. Now the entire ecosystem is thriving — companies, research centers, universities are increasingly using these models for their own tasks.

But the most interesting part is the third point about physical AI. This is no longer just text processing. Models have started to understand the physical world: biological structures and chemical processes. When they talk about physics, it’s about the ability of AI to handle the dynamics of fluids, particle physics, and even quantum physics. The dynamics of systems and the behavior of matter — the models are already catching on. It’s a completely different level than a year ago.