其次，大模型的记忆能力有缺陷：大模型在训练时“记住”了大量知识，但训练完成后并不会在使用中持续学习、“记住“新知识；每次推理时，它只能依赖有限长度的上下文窗口来“记住”当前任务的信息（不同模型有不同上限，超过窗口的内容就会被遗忘），而无法像人一样自然地维持稳定、长期的个体记忆。但在真实业务中，我们需要机器智能有强大的记忆能力，比如一个AI老师，需要持续记住学生的学习历史、薄弱环节和偏好，才能在后续的讲解与练习中真正做到“因人施教”。
客运方面，自 2 月 15 日（腊月二十八）至 2 月 23 日（正月初七），探亲流、务工流与旅游流叠加，全国铁路日均发送旅客 1341 万人次。，这一点在新收录的资料中也有详细论述
，更多细节参见新收录的资料
Leeds say they will struggle to break even on Vegas as the Super League teams have to pay all their own costs. So how difficult a decision was it to give up a home game to go? “It was a big decision and one that we didn’t take lightly. Part of our strategy is to constantly raise our profile and when you looked at the results from a marketing and audience perspective for Wigan v Warrington in Vegas last year, the eyeballs on that were incredible. You don’t get given a pot of money: you have to generate your own money through ticket sales. But like Leeds, we felt that we have a big enough fanbase to financially support our ability to go out there. It’s an incredibly tough schedule but to put ourselves on that stage was too big an opportunity to turn down. A year ago we said: ‘What if we won the Grand Final? It’ll be the World Club Challenge and straight into Vegas.’ We just decided to worry about it when it happens. And now it’s happened!”，详情可参考新收录的资料
If we have two instances of this struct and assign one to the other,