Why does AI remember some things and forget others?
Three different things people mean when they say "the AI knows".
On Monday you tell it you're vegetarian.
On Wednesday, in a new chat, it suggests a chicken recipe.
On Thursday it mentions your dog's name, which you said once, weeks ago.
So does it remember or not? There are three separate things going on, and they behave very differently.
Before it was released, the model was trained on a huge amount of text. That's where its general knowledge comes from.
Then training stops and the model is frozen. Chatting with it doesn't change it, any more than asking a printed encyclopedia a question changes what's printed inside.
Some companies can use your chats to help train future models, if your settings allow it.
That happens months later, mixed in with millions of other conversations. It isn't the AI learning about you, and it won't make the version you're using now remember you.
Context: everything the model can see in this conversation. Your messages, its replies, any files you attached.
Picture a desk. Whatever's on the desk, it can use. Whatever isn't, doesn't exist as far as it's concerned.
A new chat is a clear desk.
"Memory" features in ChatGPT and Claude work like this: the app keeps a few short notes about you. When you start a new chat, it quietly puts some of those notes on the desk before you type anything.
The model didn't remember you. The app did, and told it.
Back to the chicken: the vegetarian thing never made it into the notes. The dog did.
Your chat only ever reaches the model through the desk.
Which one is it?
If it needs to know something, put it on the desk. Paste it, attach it or say it in this chat. Don't count on memory.
Look at your memory settings once. See what it has written down about you, and delete what's wrong.
Check whether your chats are used for training. On work accounts they usually aren't. On personal accounts it's a setting you can change.