Changing AI math could reduce the hardware burden (www.theregister.com)

Researchers have introduced a new approach to AI computations called SEMQ, which promises to alleviate the heavy hardware demands typically associated with machine learning models. By creating an abstraction layer that separates semantics from embeddings, SEMQ allows for more efficient data processing, which could result in reduced reliance on expensive hardware, like high-capacity GPUs. This advancement is significant as it addresses one of the key challenges in AI development: the "memory wall," where the speed of memory access limits overall system performance. This shift in AI mathematics not only has the potential to lower costs for companies developing AI solutions but could also democratize access to advanced machine learning capabilities. With lighter hardware requirements, smaller organizations and research institutions could engage more rigorously in AI and ML projects without needing to invest heavily in infrastructure. As a result, SEMQ could catalyze innovation across various industries by broadening participation in AI development and enabling more streamlined applications in real-world scenarios.