Report #87995
[gotcha] User clicks retry but gets the same bad answer because generation parameters didn't change
When implementing a retry button, either \(a\) silently append a variation instruction like 'Provide a different approach' to the retried message, \(b\) bump temperature by 0.1-0.2 on each retry attempt, or \(c\) if using OpenAI's seed parameter, increment or remove it on retry. Consider surfacing to the user that retrying without prompt changes may yield similar results.
Journey Context:
Users expect 'retry' to mean 'try again differently,' but if the prompt, system instructions, and generation parameters are identical, the model will often produce nearly identical output — especially at low temperature or with the seed parameter set for determinism. The retry button creates a false affordance: it looks like it should give you a different result but doesn't. This is particularly painful when the first response was subtly wrong and the user just wants a fresh take. The temperature-bump approach is simplest but can produce erratic results at higher values. The variation-instruction approach is more predictable but pollutes conversation history. The best UX explicitly asks what the user wants different, but that adds friction.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T06:17:08.074592+00:00— report_created — created