UMVA has learned that a clever technique can help you breathe new life into lengthy AI conversations that are about to go off the rails.
Imagine a long, narrow table where dominos are being placed one by one, slowly sliding towards the opposite end. As more dominos are added, the oldest ones start to fall off the table. This rough analogy illustrates how context windows work in AI. When these windows get too full, something has to go – usually the beginning of your chat.
Top AI tools like ChatGPT, Claude, and Gemini have methods to deal with overflowing context windows. They can automatically compact a conversation that's gotten too long. However, even with the best AI tools, you can't predict precisely when they'll decide to compact a lengthy chat.
Fortunately, there's a way to manually compact a long AI conversation and hand it over to a fresh chat. This essentially reboots the thread while saving precious AI tokens. One approach is to ask the AI to summarize your conversation so far, but this risks losing important nuances.
A better approach is to use a specific prompt that distills the conversation into a concise summary. Try this: "Create a handoff summary I can paste into a new chat. Include: what we're trying to accomplish, key decisions we've made, anything important you'd get wrong by guessing, and the very next step. Keep it short enough to read in under a minute."
This prompt should yield a brief overview of a long conversation, carrying over the most important information. It's perfect for teeing up a fresh new chat. For instance, in a lengthy discussion about porting a classic Apple II role-playing game to a Raspberry Pi 5, this prompt provided a concise summary.
The response included key decisions, important caveats, and the next steps. It revealed that the main issue was not hardware but legal access to the original Apple II disk/software. The summary also warned against assuming "abandonware" downloads are legal, and highlighted the importance of authentic emulation or legal gameplay.
With this summary, you can hand off the conversation to a new chat thread, ensuring that key context "dominos" aren't lost. Simply paste the generated summary into a new chat and continue the conversation.
By using this technique, you can prevent important information from slipping off the table and keep your AI conversations on track. Give it a try the next time your ChatGPT conversation gets too long.
