After browsing through the technical paper of Llama 3.1, I came to the conclusion that LLMs are “Larege Luxury Models.” The cost to train these models has increased exponentially over the years and the performance seems to be peaking, it raises the question: is there a more efficient way forward for LLM innovation?