MIT Researchers Develop AI 'Telepathy' for Sharing Computational Data Between Models

Researchers at MIT have developed a new technology, dubbed "AI telepathy," that allows different artificial intelligence models to share internal computational data. This enables AI systems to leverage the strengths of various models, even if they are not designed to process the same type of information. For example, internal data from an image-processing AI like Google's Gemini 3 4B was transferred to a text-only model, IBM's Granite 3.3 8B, significantly improving the text model's ability to answer image-related questions. This could lead to more efficient AI systems that inherit prior computations when switching tasks, rather than starting from scratch.