A Russian startup called Mostik has developed a method that enables artificial intelligence models to communicate, effectively allowing them to share information without generating text output. The technique, which uses the mathematical values in model weights to facilitate interaction, allows smaller models to benefit from the capabilities of larger ones more efficiently. This approach has enabled Mostik to create a hybrid system that performs halfway between two different models, with significantly reduced costs. Source: wired
Mostik demonstrated its system by connecting two Chinese open-weight models: the 753-billion-parameter GLM-5.2 and the 4-billion-parameter Qwen-3.5. The resulting hybrid model costs one-twentieth of the full GLM model and achieves performance halfway between the two. The startup’s CEO, Sasha Malysheva, explained that the method allows models to interact without producing text output, which could improve the efficiency of running AI systems. Source: wired
Malysheva noted that combining multiple models often yields better results than relying on a single model. She emphasized that the approach could enhance the value of open-weight models, helping them compete with proprietary models from companies like Anthropic and OpenAI. The technique, she said, could also lead to the development of more specialized models in various fields such as biology and physics. Source: wired
