“The development of large language models is beginning to hit diminishing returns due to data walls and compute limits, causing their progress to follow an S-curve rather than an exponential one.”