“The future trend in large language models in the next few years will be towards Mixture-of-Experts (MoE) architectures.”