The analogy

Imagine tipping out two different jigsaw puzzles onto the same table and mixing them. Every time you look for a piece for the first picture, you risk grabbing one from the other, because they're all there, jumbled together. The more you search, the more you get confused. The only way not to go crazy is to keep the two puzzles on two separate tables.

The AI's chats are those tables. If you mix several topics in the same conversation, the AI fishes out the wrong "pieces" from the previous subject. One table per puzzle, one chat per topic: that's how the answers stay sharp.

How it really works

With every answer the AI rereads the whole conversation. If halfway through you change topic, all the text of the old subject is still there and keeps influencing the answers on the new one: the AI can drag along assumptions, tone or facts that belonged to before. On top of that, a chat that mixes many subjects becomes long and cluttered, and a bloated conversation degrades anyway, because the important details get lost in the heap. The result is an AI that seems distracted, while it's just reading a table covered in pieces from different puzzles.

What you can do in practice

  • One chat per topic or per project. It's the habit that on its own solves most of the confusion.
  • When you really change subject, open a new conversation instead of veering within the current one.
  • If you absolutely must change on the spot, state it: "let's change topic, ignore the previous context and start from here." It helps, but it isn't perfect.
  • Give a title or keep in mind which chat covers what, so you find the thread again without piling everything into one.

A common misconception

People think it's enough to tell the AI "change topic" for it to start clean. It helps, but it doesn't erase anything: the old text is still on the conversation's whiteboard and can keep influencing it, even after the notice. The only real cleanup is a new chat, where the whiteboard is truly empty. Counting on "forget what was said before" is relying on a promise the AI can't fully keep.

Frequently asked questions

Can I talk about several things in the same chat?

You can, but you risk confusion and cross-contamination between subjects. For quick, related things it's fine; when the topics are distinct and you care about the quality of the answers, separating them into different chats pays off much more.

Does telling it to forget the context work?

In part. It shifts the attention toward the new subject, but it doesn't remove the previous text, which stays readable and can still weigh in. It's a useful palliative when you don't want to open a new chat, not an erasure.

Why does it mix precisely the two subjects?

Because it has both in front of it: rereading the whole conversation, it has no real boundary between "before" and "now." It treats the whole thing as a single context, and from there the swapping of pieces between one subject and the other is born.

Isn't one "do-it-all" chat more convenient?

Convenient to open, inconvenient in the results. The more subjects you pile into a single conversation, the more the answers get contaminated and the chat gets heavier. Many focused conversations seem more cumbersome to manage, but they give you cleaner answers and let you find the right context again in a flash. The real convenience is not having to untangle two mixed puzzles every time.