“Nearly all modern large language models utilize a mixture of experts architecture.”