Remembering and forgetting with ChatGPT - A guide for beginners

1. basics of memory with ChatGPT

1.1 Cross-chat remembering and forgetting

Firstly, it is important to understand the difference between memory within an ongoing chat and memory across multiple chats.

When you talk to ChatGPT, it can remember everything that was discussed during the current chat. However, as soon as the conversation ends, this information is usually forgotten. So by default, ChatGPT is like a whiteboard that is erased with every new chat.

Only when the "memory" is activated can it remember details across several chats. You then have the option of explicitly asking ChatGPT to save certain information that is important to you.

1.2 Activation of the memory

You can activate the memory in the settings (Settings, Personalisation, "Memory" on), which enables ChatGPT to remember specific information that is relevant over a longer period of time. You will always be asked before something is memorised. You can ask ChatGPT to update or completely forget certain data at any time. This gives you control over their memories.

Example: In my conversation with ChatGPT, I told her that my name is Tara Maria Bosenick, I live in Hamburg, I'm a trans woman and I have 24 years of experience as a UX consultant. After I specifically asked her to save this information, she stored it in her memory. If I had deactivated the memory, these details would have been lost to ChatGPT as soon as the chat ended.

2. technical limitations of memory

2.1 Memory limitation during a chat

During a single chat, ChatGPT can memorise everything that is discussed, but there is a technical limit. This limit is around 8,000 tokens - this corresponds to around 6,000-6,500 words, depending on the complexity of the language. As the chat gets longer, ChatGPT gradually "forgets" the oldest parts of the conversation to make room for new information (see also my detailed blog post on this: https://www.uintent.com/de/case-studies-und-blog/everything-you-need-to-know-about-token-chatgpt).

This means that in very long conversations, older details may be lost while more recent information is still remembered. This limit serves to keep the performance of the system stable and the answers coherent.

2.2 Difference between active memory and stored information

The limit of 8,000 tokens only applies to individual chats. When persistent memory is enabled, ChatGPT can remember information between different conversations as long as you ask it to. In this case, there is no specific limit as with an ongoing chat, as saved information can be retrieved in the long term until it is manually deleted.

3. memorisable information and typical situations

3.1 Types of storable information

ChatGPT can store various types of information that go beyond simple reminders. These include:

• Personal details: Your name, place of residence, profession or other relevant facts you would like to share.

• Preferences and communication style: You can tell her whether you prefer formal or informal communication and she will adapt accordingly.

• Recurring topics or interests: If there are topics you want to talk about regularly, ChatGPT can store this information to better cater to your interests.

3.2 Typical situations for storage requests

There are certain situations where ChatGPT might suggest saving information to better help you:

• Recurring topics: If you often talk about a particular hobby, she might ask if you want her to memorise it.

• Personal details that are repeatedly relevant: If you often mention that you live in a certain city, she may ask if she should remember where you live.

• Communication preferences: If you have a preference for a certain language style, she can save this so that she can orientate herself to it in future chats.

4. control over memory

4.1 Updating and forgetting on request

You can ask ChatGPT to update or forget saved information at any time. If something changes, such as your place of residence or your professional situation, you can tell her what to adjust. You can also ask her to delete individual pieces of information or even everything she has memorised. You can manage this yourself in Settings -> Personalisation -> Manage reminder.

4.2 Transparency and insight

An important point when using the memory is transparency. You can ask ChatGPT at any time to give you an overview of what it has memorised. This way, you always stay in control and can ensure that no unwanted information is saved. You can either do this via the settings or ask them directly in a chat.

4.3 Deactivation of the memory

If you wish, you can deactivate ChatGPT's memory (Settings, Personalisation...). In this case, it forgets everything that was previously saved and you start again with a "blank page". If the memory is reactivated later, it would have to be rebuilt.

5. use of the memory

5.1 Personalisation of responses

With an activated memory, ChatGPT can adapt its answers to your specific preferences and interests. It uses the stored information to give you more targeted and relevant answers. For example, in my case, it takes into account my preferred language style - "precise, clearly formulated, structured, friendly and polite, relaxed but professional, curious and open" 😉

5.2 Limits of memory

Even though ChatGPT can save personal information, there are certain limitations. For example, it does not automatically save complete documents or longer texts from conversations. You would have to explicitly ask for certain content to be added to its memory in abbreviated form. It also only retains information that is relevant to current and future conversations. If something is no longer needed, you can ask her to delete it.

Conclusion

ChatGPT's memory function is a cool tool that helps you to receive personalised and continuous support. You have control over what information is saved and can intervene to update or delete details. The memory feature makes using ChatGPT feel more "fluid".