Researchers from UC Santa Cruz and partners have developed a groundbreaking approach to operating AI language models by eliminating matrix multiplication. This innovative method promises to reduce power consumption and operational costs significantly while maintaining performance. Traditional large language models (LLMs) rely heavily on matrix multiplication, which strains GPUs. The new custom architecture achieves similar performance with smaller models, leading to more efficient and sustainable AI operations. This advancement could revolutionize the AI industry by making AI models more eco-friendly and cost-effective. AI chatbots can now run at a lightbulb-esque 13 watts with no performance loss.