Wysor · working with AI
AI is forgetful. The fix is a new chat.
An AI has a memory, and like any memory it is finite. In a long conversation the things you said early on gradually fall out of focus, so the assistant starts to miss details and even contradict itself. People call this context rot. It is not a glitch, it is just how the memory works, and the reliable fix is refreshingly simple: start a new chat.
Before anything else: Frau Weber is allergic to penicillin.
58 more messages: symptoms, results, history, back and forth. The rule scrolls further and further out of view.
Here is the prescription: Amoxicillin 500 mg, three times daily for seven days.
Amoxicillin is a penicillin. By message 60, the note from message 1 had scrolled out of the AI’s working memory. It simply forgot.
The note never left the chat. It just fell out of what the AI could still hold in mind. People call this context rot.
It is memory, not a malfunction
You would not expect to recall every detail of a three-hour conversation perfectly either. An AI’s memory works differently from yours, but it shares one limit with it: there is only so much room. Fill a chat with enough back and forth and the early details get crowded out by everything that came after.
And it is not a quirk of one app or one model. When researchers set out to measure it, they kept finding the same thing everywhere.
75% → 55%
A fact a model answered right about three times in four when it sat at the top of a long input was found only about half the time once it was buried in the middle. Same fact, just moved.
Stanford, “lost in the middle”18 models
A 2025 report tested 18 of the leading models, including the newest from OpenAI, Anthropic and Google. Every one grew less reliable as the input got longer, even on simple tasks.
Chroma, 2025The newest models make a headline of their memory, some now advertise a million words or more. It helps, but it does not cure this: a fuller window is still a foggier one. So the fix is not more context. It is less, a shorter and cleaner conversation, which is exactly what a new chat gives the model.
Start a new chat.
When you move on to a new task, or the assistant starts to slip, open a fresh conversation. It clears the crowded memory and gives the model a clean, short context to work from. In Wysor that is one click, and your files and past chats stay safe in your workspace, so a fresh start never loses your work.
One task per chat
Keep each conversation focused instead of letting one run all day.
Put what matters first
Restate the key facts right before you ask for something important.
Upload, don’t paste
Give the AI a document it can reopen, not one long wall of text.
Common questions
What is context rot?+
It is what people call it when an AI gets forgetful in a long conversation. The further back something was said, the more likely the assistant is to lose track of it, so a long chat slowly gets less reliable.
Is the AI broken when this happens?+
No. An AI has a working memory that is finite, a bit like ours. You would not perfectly recall every detail of a three-hour conversation either. Once a chat gets long enough, the earliest details get crowded out. It is a limit of how the memory works, not a fault.
What is the fix?+
Start a new chat. A fresh conversation clears the crowded memory and gives the assistant a clean, short context to work from. It also helps to keep one task per chat, and to restate the key facts right before you ask for something important. In Wysor a new chat is one click, and your files and past chats stay in your workspace.
Fresh chat, sharp answers.
Every leading AI model, in a private workspace where a new chat is always one click away.