Researchers upend AI status quo by eliminating matrix multiplication in LLMs

This model dramatically lowers the energy cost of LLMs and approaches the same performance at large model sizes.