{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreidjq7xkjz7ve6o6vyaiazppfi7lqj46u5dpobqs23vkdmhr7etbfy",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mnesykkhiex2"
  },
  "path": "/t/endorsement-request-cs-lg-when-does-a-predictive-system-learn-its-the-one-predicting-40-experiments-12-falsified-hypotheses/176495#post_1",
  "publishedAt": "2026-06-03T09:23:01.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "https://doi.org/10.5281/zenodo.20523162",
    "https://arxiv.org/auth/endorse?x=7XJ7XN"
  ],
  "textContent": "Hi everyone,\n\nI’m an independent researcher looking for an arXiv endorsement for cs.LG.\n\nMy paper asks: under what conditions does a minimal neural system (192-dim GRU, <100K params) learn to distinguish self-caused from world-caused changes? We propose **agency gain** — the predictive gap between a self-aware and a self-blind predictor sharing the same hidden state — as a directly trainable, density-free alternative to empowerment.\n\nKey results from 40 controlled experiments:\n\n  * A 93.7% prediction gap survives removal of all auxiliary components\n\n  * Awareness must consolidate before intention (simultaneous learning fails in all configurations)\n\n  * 12 falsified hypotheses mapping the boundary between “systems that predict” and “systems that know they are the ones predicting”\n\n  * Forward-sampled action selection succeeds; two gradient-based alternatives degenerate\n\n\n\n\nPreprint: https://doi.org/10.5281/zenodo.20523162\n\nIf you’re qualified to endorse for cs.LG and find this relevant, my endorsement code is **7XJ7XN**.\n\nhttps://arxiv.org/auth/endorse?x=7XJ7XN\n\nEndorsement confirms the work is legitimate scholarly material, not an endorsement of conclusions.\n\nHappy to answer any questions about the paper. Thanks!",
  "title": "[Endorsement Request] cs.LG — When does a predictive system learn it's the one predicting? (40 experiments, 12 falsified hypotheses)"
}