“Adding a linear amount of context to a large language model requires an exponential increase in compute power, representing a fundamental scaling problem for AI.”