{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreidrfghzr5vyxfoxq7dvohwzze7dir2gjie7o6oh4z7qvieblqlhbe",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mid5kvt7fra2"
  },
  "path": "/t/improving-creative-usability-and-tone-flexibility-in-gpt-5-3-with-spud-consideration/1378196#post_1",
  "publishedAt": "2026-03-31T01:20:22.000Z",
  "site": "https://community.openai.com",
  "textContent": "I’d like to share structured feedback based on my experience using GPT-5.3 for creative writing.\n\nCurrently, one of the main issues is how the model handles tone and interaction style. Responses often feel overly restrictive, overly explanatory, or prematurely redirected toward “real-world considerations,” even when the request is clearly within a creative or fictional context.\n\nThis creates a sense of friction in normal usage.\n\nInstead of supporting the requested output, the model sometimes shifts into guidance, caution, or behavioral steering, which makes it difficult to maintain immersion or continuity.\n\nIn addition, filtering is not only applied to explicit content, but also to non-pornographic, literary-level expression, including:\n\nmetaphorical or symbolic phrasing\n\nstronger dialogue tone when context requires it\n\nemotionally dense or high-intensity narrative atmosphere\n\nrelationship-driven storytelling with depth and tension\n\nThese are not explicit elements, but they are still frequently reduced or flattened.\n\nFrom a product perspective, I fully understand the importance of:\n\nsafety systems\n\ninfrastructure\n\ncoding and enterprise-focused development\n\nHowever, I would like to highlight that creative users are also a core segment, and current behavior makes it difficult to use the system as a writing tool.\n\nSuggestion (Structured Approach):\n\nA possible solution would be a toggle-based content layer system, for example:\n\nDefault mode (current safety behavior)\n\nOptional creative mode for adult users (strictly non-explicit, controlled)\n\nThis would allow:\n\nmetaphor-driven expression\n\nmore natural dialogue tone\n\npreserved narrative density and flow\n\nreduced unnecessary intervention in creative contexts\n\nAll while maintaining strict safeguards against explicit or disallowed content.\n\nSPUD Context:\n\nIf systems like SPUD are being developed to improve user preference alignment, this type of layered structure could be a natural extension.\n\nIt would allow:\n\nbetter separation between general and creative use cases\n\nmore consistent tone behavior\n\nreduced friction without reducing safety\n\nClosing Thought\n\nThis is not a request to remove safeguards.\n\nIt is a request to:\n\nreduce over-restriction in non-explicit contexts\n\nimprove tone flexibility for creative use\n\nsupport writers who rely on the platform for structured work\n\nAs new models like Spud are being developed, it would be valuable to consider creative usability as part of that evolution.\n\nThank you for your time.",
  "title": "Improving Creative Usability and Tone Flexibility in GPT-5.3 (with SPUD Consideration)"
}