A team from Multiverse Computing in San Sebastián, Spain, led by Borja Aizpurua, has made strides in enhancing large language models (LLMs) using quantum circuits, as detailed in their recent preprint on arXiv. This innovative approach addresses the growing memory demands associated with scaling LLMs like ChatGPT and Claude, which rely on vast numbers of adjustable parameters. Instead of adding numerous classical parameters, the researchers incorporated small quantum circuit blocks into pre-trained models. This hybrid system allows for the encoding of complex relationships with fewer parameters. Testing their method on the eight-billion-parameter Llama 3.1 8B model, they achieved a 1.4% reduction in perplexity with only 6,000 additional parameters. They also observed consistent performance improvements in the smaller 135-million-parameter SmolLM2 model. While current performance gains are limited by quantum hardware capabilities, the researchers believe that advancements in quantum technology could significantly enhance AI development, offering a cost-effective alternative to expanding infrastructure.