Lesson 1041 of 1570
Cleaning the Chat When Claude or ChatGPT Gets Confused
When Claude or ChatGPT starts repeating bad answers, start a fresh chat — the context window is poisoned.
Lesson map
What this lesson covers
Learning path
The main moves in order
- 1The big idea
- 2context window
- 3fresh chat
- 4token poisoning
Concept cluster
Terms to connect while reading
Section 1
The big idea
AI chats remember everything in the current conversation. That's great — until you've gone in circles for 10 messages and the AI keeps suggesting the same wrong fix. Once a chat goes off the rails, it's faster to start a new one than to argue your way out. This is called context poisoning, and the cure is free: ⌘N.
Some examples
- Claude keeps suggesting `useState` when you need `useReducer` — you start fresh and it gets it first try.
- ChatGPT keeps including a deprecated import you mentioned 20 messages ago — new chat, problem gone.
- Cursor's chat insists on TypeScript syntax in your JS file because of an earlier confusion — clear and restart.
- Copilot Chat keeps referencing a function you renamed — fresh thread, real progress in 30 seconds.
Try it!
Next time you're three replies deep and the AI is still wrong, hit new chat. Paste only the current code and current question. Notice the speed.
Key terms in this lesson
End-of-lesson quiz
Check what stuck
15 questions · Score saves to your progress.
Tutor
Curious about “Cleaning the Chat When Claude or ChatGPT Gets Confused”?
Ask anything about this lesson. I’ll answer using just what you’re reading — short, friendly, grounded.
Progress saved locally in this browser. Sign in to sync across devices.
Related lessons
Keep going
Creators · 50 min
Long-Context Code Understanding — The 1M-Token Era
Frontier models now read a million tokens of your codebase in one shot. That changes how we architect prompts, retrieval, and the cost curve of agentic work.
Creators · 12 min
Context Rot — Why Long Sessions Get Stupid
Long agent sessions degrade in predictable ways. Learn what context rot looks like, why it happens even with million-token windows, and the compaction discipline that keeps quality high.
Creators · 35 min
AI for Debugging Stack Traces
Use AI to interpret cryptic stack traces and locate the failing line.