Consider two different systems for storing the context of a conversation for a text-generation model. System A stores the full, unaltered text of every previous turn in the conversation. System B creates a concise summary of each turn and stores only these summaries. Which system can be described as having 'adequate capacity' to represent the context, and why is the other system considered to have inadequate capacity?