New JEPA Architecture Works Across Seven Scientific Fields
Researchers extended Yann LeCun's JEPA architecture to seven scientific fields, from physics to medicine.
Researchers at PhAI Labs, working with colleagues from Stanford, Oxford, and Princeton, built an improved version of JEPA. JEPA is the world-model architecture created by AI researcher Yann LeCun. The new model can handle tasks in seven different fields, including physics, robotics, medicine, and meteorology, using just one model. In one example, the model proposed a potential treatment for liver cancer. It suggested combining the cytokine IL-18, which stimulates the immune system, with suppression of the enzyme CD73, which tumors use to weaken the immune response. Lab tests on patient tissue samples showed this combination killed tumor cells more effectively than either treatment alone. That said, this was only a lab experiment. In another test, the model estimated the exponent in Kepler's third law. It gave -1.4991, while the correct answer is -1.5.
This is interesting mainly because it shows one architecture working across fields as different as physics, medicine, and meteorology. It does not tell us much about the odds of actually developing a liver cancer drug, since this was just a lab experiment. We also do not know how accurate these models need to be before anyone can use them to design real experiments. Right now, there is more hype here than substance.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles