Researchers at the University of California, Santa Cruz have made a breakthrough by creating a large language model (LLM) running on custom hardware that only sips a mere 13 watts, which is the ...