{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreien62fqvia775syrnp3njmen2gmes23sd7kza2d37566xxck3sbuu",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mowc4l6zyrq2"
  },
  "path": "/t/feedback-for-chatgpt-improvements/1384422#post_1",
  "publishedAt": "2026-06-23T01:50:11.000Z",
  "site": "https://community.openai.com",
  "textContent": "I am a long-term user of both ChatGPT and Claude. Overall, ChatGPT already demonstrates very strong capabilities, but through extended use and more advanced applications, I have gradually observed several areas that could be further improved. If GPT-5.6 can address these aspects, they would better support advanced users, learners, and professional scenarios that require stable workflows.\n\nFirst, regarding workflow, I believe ChatGPT could benefit from adopting a system similar to Claude’s “Skills.” Claude allows users to create reusable skills that encapsulate specific instructions, processes, or capabilities, which can then be consistently invoked across different conversations. This design significantly reduces the need to repeatedly input prompts and ensures consistency in outputs. For users who need to maintain a specific style or workflow over time, such a mechanism greatly improves efficiency. Therefore, I suggest that ChatGPT provide a native skills system FOR ALL USERS in the future, allowing users to create, version, share, and apply these skills across different projects or conversations, while ensuring stability across sessions.\n\nIn terms of learning mode, although ChatGPT currently offers a Study Mode, in practice the model still tends to provide answers too quickly and lacks strict Socratic guidance. This can cause learners to passively receive information rather than actively think and reason, reducing learning effectiveness. I believe a more rigorous learning mode could be designed, where the model avoids directly giving answers and instead guides users through continuous questioning, helping them form hypotheses, reason step by step, and check understanding at each stage. At the same time, the difficulty of questions should dynamically adjust based on the learner’s level, making the experience closer to that of a real tutor rather than simply an answer provider.\n\nRegarding information presentation, I have also observed that the model sometimes expresses uncertain content with excessive confidence. This may lead users to mistake speculation for fact, thereby affecting trust. Therefore, I suggest that future versions place greater emphasis on transparency of knowledge, clearly distinguishing between facts, inferences, estimates, and opinions, and explicitly expressing uncertainty when appropriate. More frequent use of phrases such as “I’m not certain,” “this is an inference,” or “this information is unverified” can help users better understand the reliability of the information and avoid misunderstandings.\n\nAdditionally, in the experience of using custom GPTs, there are still issues with inconsistent instruction adherence. Especially in long conversations, the model may gradually drift away from its originally defined behavior. This reduces the reliability of custom GPTs and forces users to repeatedly restate their requirements. I believe future improvements could strengthen the model’s ability to consistently follow system and custom instructions, while also providing transparent explanations when it intentionally deviates from them. It could even allow users to inspect how well the model is adhering to instructions, further enhancing controllability and trust.\n\nIn terms of reasoning ability, although ChatGPT already has strong analytical capabilities, it may still exhibit logical inconsistencies or contradictions in long conversations or complex tasks. This increases the burden on users to verify results and affects overall reliability. Therefore, I suggest that future models strengthen consistency checks and contradiction detection mechanisms, performing multi-step reasoning validation before generating final answers, and proactively identifying and correcting potential errors to produce more stable and trustworthy outputs.\n\nFinally, regarding the working interface and creative workflow, I believe Claude’s Artifacts provide a highly valuable reference. It clearly separates conversation from actual outputs (such as documents, code, plans, or structured content) and offers an editable workspace, making it easier for users to iterate and refine their work. This design greatly improves efficiency and clarity for users handling complex projects over extended periods. Therefore, I suggest that ChatGPT also provide a similar native workspace in the future, supporting document editing, interactive code, project planning, version history, and side-by-side display of conversation and content, creating a more complete and professional user experience.",
  "title": "Feedback for ChatGPT Improvements"
}