Mistake L2 · Context engineering informational

When It Loops on a Wrong Fix, Reset the Context

What it is

A recovery move: when ChatGPT cycles through variations of the same wrong fix, stop feeding the failing thread and open a fresh chat with a clean, well-framed statement of the problem.

break the doom loop01Fix fails02Fix fails again03Stop — reset context04Restate the problemfresh
After a couple of failed attempts, a clean restart beats another retry.

Why it works

Once a conversation has several failed attempts in it, that history anchors the model — it keeps circling the same wrong assumption because the thread reinforces it. A reset drops the baggage and lets the model approach the problem without being primed by its own earlier mistakes.

When to use it

After roughly two or three attempts on the same bug produce minor variations that all fail. The telltale sign is the model repeating a discredited assumption.

When not to use it

When each attempt is genuinely making progress or narrowing the cause — there the accumulated context is helping, so keep going.

Prompt

[New chat] I have a bug I've been unable to fix. Fresh framing:
Symptom: <exact behaviour>.
What I've ruled out: <the failed theories, briefly>.
Minimal repro: <snippet>.

Approach this from scratch — don't assume my previous theories were right.

Example

After three rounds where ChatGPT keeps blaming the async handler, a fresh chat stating 'ruled out: async timing' leads it straight to the real culprit — a caching layer.

Common mistakes

  • Feeding 'still broken' into a poisoned thread over and over.
  • Not telling the fresh chat which theories you've already ruled out.
  • Resetting too early, before the current thread has actually stalled.

Related