{
  "id": 591213,
  "title": "Off-topic: Data Leakage and Inflated Research Results",
  "url": "/competitions/cmi-detect-behavior-with-sensor-data/discussion/591213",
  "author_name": "",
  "post_date": "2025-07-26T07:50:24.388208900Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Off-topic: I've noticed data leakage is easy to introduce when using AI for coding. Does this suggest that many published papers with high results might have achieved them through data leakage?</p>",
  "messages": [
    {
      "id": "3254295",
      "postDate": "07/26/2025 07:50:24",
      "content": "<p>Off-topic: I've noticed data leakage is easy to introduce when using AI for coding. Does this suggest that many published papers with high results might have achieved them through data leakage?</p>",
      "rawMarkdown": "Off-topic: I've noticed data leakage is easy to introduce when using AI for coding. Does this suggest that many published papers with high results might have achieved them through data leakage?",
      "votes": null
    },
    {
      "id": "3254354",
      "postDate": "07/26/2025 10:41:23",
      "content": "<p>Many forms of leakage are common in research papers for non-ml fields. I think it is less common in papers by ml dedicated researchers (they should at least know better).  As an example I recently read  review paper for deep learning in forestry and over half the reviewed papers had some form of leakage. </p>",
      "rawMarkdown": "Many forms of leakage are common in research papers for non-ml fields. I think it is less common in papers by ml dedicated researchers (they should at least know better).  As an example I recently read  review paper for deep learning in forestry and over half the reviewed papers had some form of leakage.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3254354,
      "author_name": "devinanzelmo",
      "author_url": "",
      "post_date": "07/26/2025 10:41:23",
      "content": "<p>Many forms of leakage are common in research papers for non-ml fields. I think it is less common in papers by ml dedicated researchers (they should at least know better).  As an example I recently read  review paper for deep learning in forestry and over half the reviewed papers had some form of leakage. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3254295": "Off-topic: I've noticed data leakage is easy to introduce when using AI for coding. Does this suggest that many published papers with high results might have achieved them through data leakage?",
    "3254354": "Many forms of leakage are common in research papers for non-ml fields. I think it is less common in papers by ml dedicated researchers (they should at least know better).  As an example I recently read  review paper for deep learning in forestry and over half the reviewed papers had some form of leakage."
  },
  "source": "meta"
}