Hacker News AI· jojo960527·· 15 hr agoAI score38
We let our AI save memories by itself. What it got wrong in the first two days
We let our AI save memories by itself. What it got wrong in the first two days
AI summary
NEXUPIRA describes replacing an unused memory-approval queue with automatic saving of facts from owners' own messages, governed by code rules that drop credentials, personal data, hypotheticals, post material and the assistant's own words. The post details six failures found in two days of production testing, including a one-off request saved as a standing rule and a test question saved as a business plan, and the fixes for each. Every save now appears under the answer with an Undo button, and contradictions prompt an Update or Keep current choice in the chat.
Source: Hacker News AI · nexupira.com