Russian researchers develop AI communication method without text

Russian mathematicians at Mostik, a startup named for its bridging approach, have developed a method allowing AI models to communicate through mathematical values in their weights, bypassing text. The technique enables smaller models to leverage the capabilities of larger ones more efficiently. Mostik’s approach was tested by connecting a 753-billion-parameter GLM-5.2 model with a 4-billion-parameter Qwen-3.5 model, resulting in a hybrid system that costs one-twentieth of the full GLM model and performs halfway between the two. Sasha Malysheva, CEO of Mostik, noted that combining models often yields better results than individual ones, comparing it to estimating a pig’s weight through collective guesses. Vladimir Arustamian of Lovable praised the method, calling it a breakthrough that could enable specialized models to be trained more effectively. Karl Tuyls, a former Google DeepMind scientist, highlighted the efficiency gains of the approach, while Stanislav Smirnov, a Fields Medalist, suggested the method could reveal new insights into AI and human cognition. Malysheva, who discovered her math talent after being challenged by her brother, aims to prove skeptics wrong by demonstrating the viability of the bridge approach.