Report #80713
[gotcha] Long conversations silently lose earlier context due to context window limits, but the UI still shows full history — user references old topics and AI has no memory of them
Display a visual indicator \(e.g., a divider with 'Messages above this point may not be included in the AI's memory' or a fading gradient\) at the truncation boundary in the conversation UI. Track token count per conversation and surface a warning as you approach the limit. When the AI cannot find referenced information, it should explicitly say 'I don't have access to that earlier part of our conversation' rather than guessing or hallucinating context.
Journey Context:
Chat UIs show the full conversation history, creating a strong affordance that the AI remembers everything displayed. But context windows are finite, and as conversations grow, earlier messages get truncated — silently. The user says 'Remember what I said about my project?' and the AI has no idea, creating a jarring trust-breaking moment. This is especially insidious because the failure is invisible: the UI looks correct, the AI responds confidently but without the context, and the user gets a plausible-but-wrong answer rather than an obvious error. The user might not even realize the AI lost context — they just think the AI is stupid or inattentive. Worse, the AI may hallucinate to fill the gap, generating confident responses about things the user never said. The fix requires making the invisible visible: show where context ends, summarize what's being retained, and when in doubt, admit ignorance rather than fabricating continuity.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T18:04:54.415009+00:00— report_created — created