{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreib6dtoxonqpjplamctburnrmzsftgahikxseatchoteayxfumtqvm",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mpl6eietbov2"
  },
  "path": "/t/llm-curving-via-prompting/177166#post_17",
  "publishedAt": "2026-07-01T07:28:30.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "Colab"
  ],
  "textContent": "I think the problem is in linear probabilistic trajectory enforced by its creators. As you mentioned there are data of everything and if you don’t let so connect to all the data it can predict only the linear trajectory.\n\nWhat I am prosposing in my work insteda is to seek density and drifts in it that will show you density is neccessary for more accurate answers for those who don’t see the world blindly reduced to nothin else than predefine linear trajectoy.\n\nThe metrics mentioned in upper post [Colab] is actually not measuring linear trajectories but expose field condition in which or because of which semantic drift happens.",
  "title": "LLM \"curving\" via prompting"
}