A simple physics-inspired model sheds light on how AI learns

[Binary_Bark]

Artificial intelligence systems based on neural networks—such as ChatGPT, Claude, DeepSeek or Gemini—are extraordinarily powerful, yet their internal workings remain largely a "black box." To better understand how these systems produce their responses, a group of physicists at Harvard University has developed a simplified mathematical model of learning in neural networks that can be analyzed mathematically using the tools of statistical physics.

https://techxplore.com/news/2026-05-simple-physics-ai.html