{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreic34rdwif6l577bhqhpm2eolextobvfsum52wcokwcor3zkj6migi",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mlw5nwco6ob2"
  },
  "path": "/t/physical-ai-safety-ownership-and-execution-boundaries/175776#post_10",
  "publishedAt": "2026-05-15T18:49:25.000Z",
  "site": "https://discuss.huggingface.co",
  "textContent": "Jang-woo:\n\n> Before answering or executing, is there enough information?\n\nthis is actually a painful; question to ask because… how does the AI KNOW if there is not enough information?\n\nif you look at what AI is trained on, which is at minimum billions of different paremeters, the AI has alot of information to compare against that it brings to the table. and i think this is 1 part of the equation, and what this implies is that the AI is mistaking the information it already has, as the completion for the information that was given.\n\nthe other part… lack of information identification.\nif we consider an idea introduced as a shape, lack of information produces a shape with 3 - 8 sides.\n\nbut comprehensive information _produces a shape_ that has a minimum of 50 sides.\n\nperhaps the agent should be expecting a shape with a minimum of 50 sides, and should query if prompt information given < 50 shape sides.",
  "title": "Physical AI Safety: Ownership and Execution Boundaries"
}