Some iteration converges and some polishes its own output while appearing to improve. The difference is whether new information enters on each pass — and the four loops sorted by which it is.
- It will find things, because it will always find things — but the critique is generated from the same picture that produced the error.
A "loop" here just means going back and forth with an AI to improve something.
The useful test: are you adding anything new each time you go round? If you bring a fact, a correction, or a real result, it gets better. If you just keep asking again, it is rearranging the same material and it will drift.
Some loops improve with each pass. Others get worse while appearing to improve, and the difference is whether new information enters on each turn.
Does anything from outside the conversation arrive on this pass? A source, a measurement, a human judgement, a real result. If yes, the loop can converge. If no, it is polishing its own output and calling it progress.
Four loops, two of which degrade
Refinement — converges
You judge the output against something real and say what is wrong. Your judgement is the new information. This is the loop that works, and it works because you are in it.
Grounded iteration — converges
Each pass runs against a test, a dataset, a rendered result. Reality supplies the correction. This is why coding loops with a test suite behave far better than writing loops without one.
Self-critique — degrades
Asking a model to review its own output. It will find things, because it will always find things — but the critique is generated from the same picture that produced the error.
It helps for format and structure. It does not help for factual correctness, because nothing new arrived.
Reprompting the same request — degrades
Not liking an output and asking again, slightly differently. Each pass carries the previous attempts in context, so the conversation is now anchored on the thing you rejected.
Worse: pushing back can produce a revised answer that is less correct. Sycophantic revision from a right answer to a wrong one occurred in 14.66% of tested cases.
The same study cuts the other way too, and this page previously left that out. Revision toward a correct answer — progressive sycophancy — occurred in 43.52% of cases, roughly three times more often. So reprompting is not simply destructive: on balance it corrected more than it broke. What it is not is reliable. Roughly one revision in seven moved a right answer to a wrong one, and nothing in the exchange tells you which kind you just got — the model sounds equally confident either way. That is the case against leaning on it, not a claim that it always degrades.SycEval, arXiv 2502.08177 — regressive 14.66%, progressive 43.52%. Both figures added 25 Aug 2026; an earlier version of this page cited only the first. More on how AI influences you back.
The compounding version
The same failure at scale: output becomes input becomes output. A summary of a summary, a rewrite of a rewrite, a dataset containing yesterday's generations.
Every pass without outside information moves the result toward the model's average and away from the specific thing you wanted.
This is why long chains of AI steps drift toward generic — not because any single step failed, but because specificity is the first thing lost when nothing replenishes it.
Working rules
- Start fresh rather than reprompting. A new conversation has no rejected attempts anchoring it.
- Bring something in on every pass — a source, a constraint, a real result. If you cannot, stop looping.
- Cap the passes before you begin. Three is usually the point of diminishing return; past that you are polishing.
- Ask it to check, not to improve. Checking is the easier task by roughly a factor of two and is less prone to drift.Kalai, Nachum, Vempala & Zhang, OpenAI, Sep 2025 — a formal bound
- Judge against the brief, not the previous version. Comparing to the last attempt guarantees you converge on the last attempt.