When you talk to an AI chatbot, it reads the whole conversation each time you send a message. All your messages, all its replies. That full conversation is called the context.
But the context has a size limit. Think of it like a window that can only show so many messages at once. As the chat grows, the oldest messages slide out of view.
This is Maria. Maria's doctor suggested she try to get out of the house more often, so she chats with an AI about her weekly routine. Early on, Maria mentions when she takes her medicine. Press play and keep your eye on what happens to that important piece of information.
The AI wasn't trying to trick Maria. It lost the real answer because the conversation got too long. The fact about her medicine slid out of its memory window. When she asked about it, the AI filled the gap with a guess that sounded confident.
People call these AI errors hallucinations. It sounds dramatic, but it just means the AI produced something that is not true. It happens more in long conversations because more early facts have been lost.
AI chatbots always write in a confident tone. They do not say "I forgot" or "I am not sure." They build each answer word by word, picking whatever sounds most natural next. A wrong guess can sound exactly the same as a correct answer.
When a conversation gets long, open a new one. Paste in any facts the AI needs to know. A shorter chat means fewer forgotten details.
If the AI gives you a fact you shared earlier, verify it against your own notes or records. Especially for health, money, or appointment details.
The AI will never tell you it forgot something. If an answer about an old topic comes back sounding slightly different from what you remember, trust your own memory first.
A long chat can lose a fact you gave it. The next lesson covers a different problem: facts the AI never had in the first place. Continue to The empty shelf.