{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreicsfwlpaw6lg24wc25xarxcozdb6zat4jlfnr3kxohtt3xoy3k7yi",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mh53h5pefnn2"
  },
  "path": "/t/how-deot-makes-llms-think-a-new-framework-for-open-ended-reasoning/174282#post_1",
  "publishedAt": "2026-03-15T13:25:00.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "How DEoT Makes LLMs Think: A New Framework for Open-Ended Reasoning"
  ],
  "textContent": "LLMs struggle with open-ended reasoning. Most frameworks focus on task completion or next-token prediction—but fall short when it comes to multi-perspective analysis or structured exploration.\nTo address this, our team at NeuroWatt developed DEoT: Dual Engines of Thoughts, a reasoning framework that decomposes complex queries into parallelizable and verifiable reasoning paths.\n\nDEoT introduces two key components:\n\n  1. Breadth Engine – surfaces diverse dimensions of a problem space\n  2. Depth Engine – performs focused, multi-step analysis within each dimension\n\n\n\nThrough this architecture, we aim to strengthen one of the current limitations of LLMs—structured associative thinking and coherent reasoning chains. Instead of relying solely on statistical completion, DEoT encourages deliberate, agent-style thought processes that can be audited, scaled, and reused.\nIf you’re building cognitive agents, financial LLMs, or any domain requiring structured thinking beyond deterministic outputs.\n\nHow DEoT Makes LLMs Think: A New Framework for Open-Ended Reasoning",
  "title": "How DEoT Makes LLMs Think: A New Framework for Open-Ended Reasoning"
}