Every single large language model only has a certain number of memory, so it may only settle for a particular range