You will be able to recognise when a conversation or document is too long for good answers and what to do about it.
You have been working on your CV with an assistant all evening. In your first message you set two rules: keep it to one page, and leave out the two years you spent at a start-up that folded. For the first hour it followed both. Now, forty messages later, the draft is a page and a half and the start-up is back in, described in glowing terms.
You did not change your mind, and the assistant did not decide to ignore you. Your first message has, in effect, faded from view. This lesson explains why that happens and what to do about it.
Lesson 5.1 described the context window as everything the model can see at once. Every message you send and every reply it writes adds to it. A long chat with pasted documents and detailed replies can fill the window faster than you would expect.
Once the conversation is bigger than the window, the app has to do something with the overflow. Products handle it differently. Some drop the oldest messages, so the model simply stops seeing them. Some replace older parts with a short summary written by the model, which keeps the gist but loses detail. Some tell you the chat is full and ask you to start a new one.
The first two are the risky ones, because they happen quietly. You can still scroll up and see your first message on screen, so it feels as if the assistant can see it too. It may not. The instruction you gave at the start is exactly the kind of thing that gets dropped, or squeezed into a summary line that leaves out "and not the start-up".
There is a second problem, and it applies even when everything fits. A model does not give every part of a long text the same weight. In practice, models tend to handle the start and the end of long material better than the middle, and a detail buried halfway through a long report or transcript is the one most likely to be missed.
Picture an operations executive at a Jurong logistics firm who uploads a fifty-page vendor contract and asks whether there is a penalty for late delivery. The penalty clause is on page 27. The assistant summarises the payment terms from the opening pages and the termination terms from the end, and says nothing about penalties, or says there are none. The clause was in the window. The model just did not use it.
So a large window helps you get long material in, but it does not promise that every line gets read with care. Lesson 5.4 has you test this on a real document.
You rarely get a warning, so learn to spot the symptoms. Three are common.
A constraint you set early stops being followed: the word limit, the tone, the thing you said to leave out, the format you asked for. A mistake you already corrected comes back, such as the wrong client name or a figure you fixed an hour ago. Answers about a file ignore part of it, and questions about a section in the middle get vague or wrong replies while questions about the opening pages are fine.
Any one of these is a signal to stop pushing the same chat further. Repeating the instruction can help for a message or two, but in a long chat the same thing tends to slip again.
The usual fix is simple. Open a new chat and give it a short summary of what matters. Write it yourself, or ask the old chat to draft one and then check it, since its own summary may already have lost the early rules. A good carry-over note has three parts: the goal, the rules that must hold, and the current state of the work.
For the CV, that might be four lines. I am applying for a regional marketing manager role. Keep it to one page. Leave out the 2021 to 2023 start-up role entirely. Here is the latest draft. The new chat starts with all of that in plain view, near the top, where it gets attention.
For long documents, split the work. Rather than uploading the whole contract and asking everything at once, give the assistant the section that matters, or work through the document a part at a time and ask the same question of each. If you need an answer that draws on the whole document, ask it to point to the page or quote the passage it used, so you can see whether it actually read the part you care about.
A useful rule of thumb: when a chat starts to feel like it is arguing with you, it is often just that the part where you agreed on the rules has gone. Most people have at least one old chat like that sitting in their history. Go and find yours, because looking at a real one shows you where your own instructions started to slip.
Find a long chat in your history, scroll to an early instruction, and check whether the latest replies still follow it; write down what you would carry into a fresh chat.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).