In China, a group of researchers from Tsinghua University and Peking University has introduced PhyE2E — an artificial intelligence system capable of independently deriving physical equations from raw data, without prompts or human interpretation.
Unlike conventional neural networks, which merely fit models to statistical correlations, PhyE2E does something qualitatively different: it uncovers the structure of reality. Trained on empirical datasets and known physical laws, this AI uses a hybrid of transformer architecture and symbolic reasoning to generate equations in the form a theoretical physicist might write.
The results astonished even the creators. When tested on NASA astrophysical data, the system not only reproduced known regularities but also proposed an improved mathematical formula for solar cycles — more accurate and elegant than previous models. In other words, artificial intelligence is not merely reproducing human discoveries — it is beginning to make its own.
PhyE2E is not ChatGPT in a lab coat, but a prototype of what physicists call “automated discovery”: an AI that learns to perceive the hidden geometry of the Universe and express it in the language of mathematics. Its equations are readable, verifiable, and — most importantly — logically interconnected.
If today’s language models have learned to write texts, then PhyE2E takes the next step — it writes physics. This is not just a tool for data analysis, but a new form of cognition where the boundary between computation and discovery begins to dissolve.
More details: www.nature.com
