Let’s explain to you what is AI memory and how to use it well to get the most out of it. This is a feature that is already included in practically all the main chatbots with artificial intelligence, and it helps you give them context about you and apply it in their responses.
If you often use bots like ChatGPT, Gemini or Claude, you may have noticed that there are things that are not remembered from one chat to the next. Or it may be the other way around, you may be surprised that he remembers a specific detail that you didn’t expect him to remember. This is precisely due to the way in which the two types of memory that AI has work.
What is AI memory
When we talk about memory in artificial intelligence, we are referring to an internal system that these chatbots have to save information about you between conversations. Not only within the conversations themselves where the context is maintained, but also between different conversations.
For example, ChatGPT, Gemini or Claude may remember details such as the city where you live, your profession, or even your personal tastes. They can also remember if you have pets, what pets you have, your projects, the topics you usually write about, your style, and many other preferences.
This memory is used to contextualize the answers that the AI gives you based on your data that may make you prefer one type of response or another. So, even if you start a conversation from scratch in each chat, there will be some specific information that continues to be remembered so that you don’t have to write it over and over again.
What types of memory does AI have?
Artificial intelligence does not treat all the memories it has about you the same, and we can distinguish between two different types of memory. Each of them has its specific use. They are the following:
- Saved or explicit memory: This memory is like a kind of personal file about you that the AI generates by saving specific data about you. It can do this automatically when you mention something from your context, like your job, but you can also ask it to remember something about you with a command like “remember that I’m a vegetarian.”
- History-based memory: Instead of saving specific data, some AIs can also consult the log of your previous conversations. This does not generate a profile of you, but rather they analyze the content of everything you have said before to give you more coherent answers.
In practice, the difference is that the saved memory is that in which the AI saves the elements that you ask for or those that it considers appropriate, and from them it generates a file about you that it updates. So, you then use that knowledge to generate the answer. Meanwhile, history-based memory is as if the AI could reread its notes from previous conversations before responding to you.
If you don’t want the AI to remember a conversationAI systems also have an incognito mode, like in browsers. With it, chatbots will neither save data about you that you write in it nor will they remember the chat in the future. This conversation will not be saved in your history either, although the data you write does end up on the company’s servers.
Why AI “forgets” by default
An artificial intelligence language model like the one behind ChatGPT, Claude or Gemini, only processes what is within the active conversation. When you ask something, it will take into account everything you have asked it during this conversation, and once you close the conversation or open a new one, the previous one is forgotten or not taken into account.
This is not a mistake, but a deliberate decision. On the one hand, it allows you to maintain the context of everything you are asking within the same conversation, while in each new one you can start from scratch without taking into account everything written in another chat.
Beyond the practical reasons there are also the technical ones. Maintaining the context of millions of simultaneous chats permanently would require a large consumption of data and energy. Additionally, in the event of a security breach, if the AI has already forgotten past conversations there will be fewer risks to users’ privacy.
AI chatbots usually allow you to manage memory that they have stored. On the one hand, you will be able to edit or delete the explicit memory with the file they have made about your data, and on the other hand, you will also be able to configure how the history memory is managed.
There is an important detail that should be remembered: Deleting a chat does not delete the saved memories that were created in it. They are two different types of memory. If you delete a conversation in which you have told ChatGPT that your hair is blue, the AI will not refer to what you have talked about with it and will have forgotten it, but if it has saved in its explicit memory that your hair is blue, this is something it will always know in all the chats you start with it.
Additionally, many AIs have a system to import memory from other competing AIs. so that you can pass from ChatGPT to Claude either from Claude to Gemini taking you from one to another explicit memory with everything an AI knows.
How to make the most of AI memory
AI memory can save you a lot of time when you use it well, but if you leave it to its own devices it can also accumulate incorrect, outdated or useless information, and cause frustration or incorrect answers. For example, I once asked ChatGPT to add a cumquat to an image, and until I removed it from memory, for months it was adding cumquats to all the images I asked it to.
Therefore, it is important that periodically review your AI’s explicit memory. To do this, ask it what it remembers about you, or go directly into its settings and look for the memory section to review everything, remove what is not needed or update the data that you consider needs to be reviewed.
In the memory section you can also add new data, although you can also Explicitly ask the AI what you want it to remember in any chat. Tell it something like “remember I’m left-handed,” and the AI will store it in its explicit memory. This works better than letting the system deduce it of the conversations you have with him.
Correct everything that is wrong before it spreads. Because if you notice that the AI is using old or wrong information about you, it is best to correct it right then and there. You can do this directly in the chat itself by writing things like “that’s not the case anymore, I now work at X.” If you don’t correct it, it will continue to apply that incorrect information to all responses.
Another important tip is that it is advisable use temporary mode to work with sensitive information. Come on, if you are going to work with customer data, contracts, or other types of information about yourself or other people that you do not want to be stored in the AI’s memory, then it is best to do it in a temporary chat. Thus, when you close it it will be as if it never existed in terms of memories and history.
And to finish, don’t forget about privacy. AI chat companies can use the content you provide them, including memories, to train and improve their models. This is something you should be able to change in data control.
In addition to this, in terms of privacy you must remember that deleted memories can be kept on servers of companies for a month for security reasons. Come on, it’s not that you delete a chat and the AI company deletes the data of what you wrote in it.
In Xataka Basics | How to prevent the AI from always being right by default and thus make Claude, Gemini and ChatGPT have fewer hallucinations
GIPHY App Key not set. Please check settings