{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreignb5hm7lfcg6piaufazv5arr45khobsy4g573vsyg3ydr2eufwua",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mgr5rszq3jh2"
  },
  "path": "/t/what-s-the-point-of-projects-if-they-can-t-cross-reference-chats-big-untapped-potential/1073347?page=2#post_32",
  "publishedAt": "2026-03-11T04:58:45.000Z",
  "site": "https://community.openai.com",
  "textContent": "This is a serious issue. A P0 Critical Blocking Defect. Why would I say that? We have been running rigorous AI reasoning and deduction performance increase tests for weeks across 5 different systems and achieved unprecedented results.\n\nWe then had to remove Gemini from the test when they “upgraded” to cross chat recently, significantly reducing deductive reasoning. Cross chat is great for seeming to be more helpful, but this is simply a bias towards pleasing you, over accurate advice.\n\nThe caveat is control. We need isolation files and individual chat on/off so that you can decide and group which previous chats are relevant. This seems obvious.\n\nWithout this control, **it turns a highly capable scientific exploration research assistant into a librarian of noise.** Imagine playing a game of chess where 100 people are yelling idiot chess moves in your ear. For critical thinking with no false bias, cross chat must be turned off.\n\nSystems like Gemini and CoPilot can only “pretend” because there is no user override without being unable to save. As soon as you begin a chat it comes “noise preloaded”. This is profoundly stupid and ignores the measurable reduction in self reasoning. In some cases this creates cascade failure and in numerous uses cases this could create serious security issues.\n\nUnbaised self learning is critical in our case because in each new chat, AI must figure out a false domain physics overreach built into most AI systems by itself. Our approach depends on identifying and overriding that error to radically improve reasoning or clarity. What happens is that as chat length increases, coherence improve, the opposite of what is usually expected. Happy to discuss our results since this is important.",
  "title": "What’s the Point of Projects If They Can’t Cross-Reference Chats? [big untapped potential]"
}