The experiment highlighted an aspect unknown to many users, which is that there are internal limits related not only to the number of messages, but also to the model’s ability to remember the entire context of the conversation.
Forget the dialogue
“ChatGPT” relies on what is known as the context window, which is the amount of information that the model can keep in its memory during the conversation. This window is measured in tokens, which are small pieces of words.
As the conversation goes on, the model gradually begins to forget old parts of the dialogue to make room for new content.
This does not mean losing the entire chat, but rather losing parts of the previous context. Modern models of “ChatGPT” can handle approximately 128 thousand “tokens” in some versions, and this is approximately equivalent to about 80 to 100 thousand words in English or less in Arabic depending on the writing style, because Arabic may need more “tokens” for each word.
According to the report, there is another type of restriction that is not clearly announced, which is the maximum conversation length.
When you access it, a message may appear stating that the conversation has reached its end and cannot be continued. This limit is different from the memory problem, because it is not related to what the model remembers, but rather to completely closing the session itself. When these limits are approached or exceeded, things begin to gradually progress instead of suddenly stopping.
The process of gradually losing old messages within the conversation may begin, and the coherence of long answers will also weaken. In some cases, closing the conversation permanently with an end message.
To avoid all of this, experts advise summarizing the conversation periodically, starting a new conversation when needed, and creating a context summary that can be copied to a new chat to complete the work. This method helps maintain continuity without losing important information.
This experience demonstrates that “ChatGPT” is not a borderless system, but rather operates within technical limitations related to memory and the length of the conversation.
Although these limits may not be noticed by the average user, they become apparent in long or complex uses. A good understanding of these limitations helps users make better use of the model, especially in extended projects or ongoing discussions that need periodic organization and summarization.
