{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreifub3z6awnb65wgthmr6gtvd4azu73y5lzmyfc6xpl3g4ushtn65u",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mnqwb3tzorb2"
  },
  "path": "/t/dreaming-v3-sacrifices-real-personalization-for-compute-savings-a-power-user-s-perspective/1383010#post_1",
  "publishedAt": "2026-06-08T03:51:03.000Z",
  "site": "https://community.openai.com",
  "textContent": "Dear OpenAI team,I’m a Pro subscriber. I’ve been using ChatGPT daily for over a year, and I relied heavily on the Saved Memories system to maintain continuity across hundreds of conversations.\nSince the June 4 rollout of Dreaming V3, here’s what changed for me in practice:\nWhat was lost\nUnder the old system, Saved Memories functioned like a notebook I controlled. I could see everything stored. I could add, edit, and delete entries. If something was in the notebook, I could be confident ChatGPT would reference it. It was simple, transparent, and reliable.\nUnder Dreaming V3, that notebook is gone. The system now decides what’s important on my behalf — and it gets it wrong. Memories I consider critical require repeated reminders, which never happened before. Meanwhile, outdated information from months ago resurfaces unprompted in the memory summary, because the system has no way of knowing what matters to me.\nThe core problem: removing user agency\nThe old system respected a basic principle: the user decides what’s worth remembering. Dreaming V3 reverses this. A background process now synthesizes and curates memories without user input, replacing explicit storage with inferred relevance.\nThis is not smarter memory. This is the system overriding the user’s judgment about their own context.\nFor users with straightforward needs — casual conversation, general Q&A — this may be invisible. But for power users who maintained structured, intentional memory states, it’s a serious loss of control.\nUnder Saved Memories, I could guarantee these facts persisted. Under Dreaming V3, the system may decide any of them are “stale” and revise or drop them without notice. There is no way to pin a memory as permanent. There is no way to mark something as “never auto-update this.”\n“Cost reduction” is not “improvement”\nOpenAI has stated that compute costs for Dreaming were reduced by ~5x, enabling free-tier rollout. I understand the business logic. But from a user experience perspective, what happened is clear: the system became cheaper to run by becoming less precise, less controllable, and less personalized.\nThis is cost reduction marketed as a feature upgrade. The old system was more expensive but gave users real control. The new system is cheaper but substitutes the system’s judgment for the user’s.\nWhat I’m asking for\n1.Bring back pinned memories. Let users mark specific memories as permanent and exempt from automatic revision.\n2.Provide an opt-out for auto-curation. Some users want to manage their own memory state. Let them.\n3. Show revision history. If the system changes a memory, show what it was before and why it changed.\n4.Don’t remove working features to subsidize free-tier expansion. Pro users are paying $100 or $200/month. We should not lose functionality so the system can scale to free users.\nThe old Saved Memories system wasn’t perfect, but it did one thing right: it trusted the user. Dreaming V3 does not.",
  "title": "Dreaming V3 Sacrifices Real Personalization for Compute Savings — A Power User’s Perspective"
}