{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreihf4ypcmkaozgrfvsjmokth6hmkz52opjhrri5ildqbamyrt45yzy",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mkqirmtll4c2"
  },
  "path": "/t/humans-see-it-instantly-llm-doesn-t/174765#post_5",
  "publishedAt": "2026-04-30T19:01:38.000Z",
  "site": "https://discuss.huggingface.co",
  "textContent": "this whole conversation points out 2 things.\n\n  1. Most AI are trained on text first, and everything else second. thats mainly a momentium problem.\n  2. instructions sets are important.\n\n\n\nwhile the goal is to get the AI to ‘solve the maze correctly’ you cannot actualy directly instruct it on how to go about doing it.\n\nyou must introduce the maze, and a prompt outlining a proposed approach.\n“Here i introduce a maze. determine if a soltuin exists, then explain your reasoning.”\n\nso, on some AI there is a setting to ‘show thoughts’ and unless this is turned on, you may not even see what thought process the AI is useing to even Approach this problem. why does that matter?\nbecause you are trying to determine whether it is pathing first, or reachability testing.\n\nin many cases, this only actually maters beyond idle curiosity if you are in research, OR you are doing active work that requires an AI to bebale to do this consistantly.",
  "title": "Humans see it instantly — LLM doesn’t"
}