{"slug": "same-net-same-seed-100-vs-50-on-unseen-magnitudes", "title": "Same net, same seed – 100% vs. 50% on unseen magnitudes", "summary": "Structured Intelligence, a company with a patent-pending encoding method, reports that its structured place-value input encoding achieves 100.0% accuracy on unseen magnitudes in Task A, compared to 52.7% for one-hot and 50.0% for learned embeddings, using identical networks, data, and random seeds. The company claims its method generalizes to completely unseen inputs without retraining, citing a held-out comparison where structured encoding reaches 98.6% unseen accuracy versus 0.65% for one-hot and 0.9% for learned embeddings. Structured Intelligence is offering a public benchmark, API access, and licensing under NDA, with Dr. Jarrod S. Segura as the contact.", "body_md": "Structured Intelligence · patent pending\n\n# Watch their table go blind.\n\nBillionaire-backed embeddings buy a slot per token. Ask a magnitude the net never saw and the slot is chance. Place value already has the column. Identical network. Identical data. Identical random seeds. The sole variable is the input encoding.\n\nTask A · unseen\n\nStructured\n\n100.0%\n\ntrain 100.0% · h=48\n\nTask A · unseen\n\nOne-hot\n\n52.7%\n\ntrain 99.2% · h=16\n\nTask A · unseen\n\nEmbedding\n\n50.0%\n\ntrain 100.0% · h=16\n\n01\n\nRun it\n\nSame net, same seed. Only the encoding changes. Unseen magnitudes expose the slot. You do not take our word.\n\n02\n\nTry it\n\nSign in, mint a key, POST integers. Public place-value rungs. No charge to try.\n\n03\n\nBuy it\n\nLicense the API, purchase, or stake. The composed family stays under NDA. You are not buying a PDF.\n\nProduct factory\n\n## Apps, not source\n\nCounsel holds the closed family. The factory ships products that sit on it: a saturation clock, success mirrors, a cockpit. Clients pay for the run. They never see the codes.\n\n[Open the factory](/products)\n\nLive miniature · seed 7\n\n## Task A — is this reading odd?\n\nTaught on 0–127. Asked on 512–1,023 — never seen. Same logistic readout. The only change is the input encoding.\n\nStructured\n\nTrain 0–127\n\nUnseen 512–1,023\n\nOne-hot\n\nTrain 0–127\n\nUnseen 512–1,023\n\nEmbedding\n\nTrain 0–127\n\nUnseen 512–1,023\n\nA slot that was never trained holds whatever it was born with.\n\nFull protocol · measured\n\n## Held-out comparison\n\nBoth structured and one-hot reach 100% on the training set. Training-set performance is not the claim. Generalization to completely unseen inputs is. That is why you pick this over a table with a billion-dollar budget.\n\n| Encoding | Seen | Unseen | Parameters |\n|---|---|---|---|\n| Structured (fixed place value) | 100.0% | 98.6% | 1,673 |\n| Learned embedding | 37.1% → 100% | 1.7% → 0.9% | 3,721 → 51,721 |\n| One-hot | 100.0% | 0.65% | 33,417 |\n\nEmbedding seen accuracy rises only after capacity is grown; unseen accuracy falls as parameters increase.\n\nTask A · measured\n\n### Odd reading, unseen magnitude\n\n- Structuredunseen 100.0%\n- One-hotunseen 52.7%\n- Embeddingunseen 50.0%\n\nTask B · measured\n\n### Add two readings, 32× jump\n\nTaught 0–15. Asked 128–511. Composed rungs trained nothing new — same primitives, one column wider.\n\n- COMPOSED (place-value rungs)100.0%\n- Structured, monolithic0.0%\n- One-hot, monolithic0.0%\n\nLive compose check on this page: 200 of 200 (place-value add, no retraining).\n\nFor a training run\n\n## Train it to know. Not to guess.\n\nToday you train a model to guess. Show it enough examples and hope the next one looks like the last. When the next one is a size it rarely saw, it guesses — that is the 50% on this page. Chance in a nice jacket.\n\nThis is training to know. Unseen is still a thing. Same net, same seeds: 100% on a number it never trained. They guess. This knows.\n\n## What the rivals need\n\nAtoms in the observable universe: ~10⁸⁰. A place-value code needs 12, 16, 32, 64, and 256 inputs.\n\n| Range | Values | One-hot slots | Place-value inputs |\n|---|---|---|---|\n| 12-bit | 4,096 | 2¹² | 12 |\n| 16-bit | 65,536 | 2¹⁶ | 16 |\n| 32-bit | 4.29 billion | 2³² | 32 |\n| 64-bit | 1.8×10¹⁹ | 2⁶⁴ | 64 |\n| 256-bit | 1.16×10⁷⁷ | 2²⁵⁶ | 256 |\n\n## Try. Then buy.\n\nThe result is public. The family is not. Run the bench, mint a research key, POST integers. If you want the API in production, a license, purchase, or a stake — that is a named conversation, not a download.\n\nBuy · named evaluator\n\nLicense, purchase, or stake\n\nCommercial use and the full encoder. NDA. You are buying a project that runs on the codes, not the codes.\n\n[Email Dr. Jarrod S. Segura](/cdn-cgi/l/email-protection#5b313a2929343f283e2f331b33342f363a32377538343664282e39313e382f66092e353c7e696b7e1e697e636b7e626f7e696b3732383e35283e7e696b34297e696b282f3a303e7d3a362b6039343f2266127e696b293a357e696b2f333e7e696b37322d3e7e696b393e353833757e696b127e696b2c3a352f7e696b2f347e696b2f3a37307e696b3732383e35283e7e69187e696b2b2e2938333a283e7e69187e696b34297e696b282f3a303e757e6b1a7e6b1a153a363e7e681a7e6b1a1834362b3a35227e681a7e6b1a)\n\nDr. Jarrod S. Segura", "url": "https://wpnews.pro/news/same-net-same-seed-100-vs-50-on-unseen-magnitudes", "canonical_source": "https://structured-intelligence.grok.me", "published_at": "2026-08-25 21:38:05+00:00", "updated_at": "2026-08-25 21:46:06.896273+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "ai-research"], "entities": ["Structured Intelligence", "Dr. Jarrod S. Segura"], "alternates": {"html": "https://wpnews.pro/news/same-net-same-seed-100-vs-50-on-unseen-magnitudes", "markdown": "https://wpnews.pro/news/same-net-same-seed-100-vs-50-on-unseen-magnitudes.md", "text": "https://wpnews.pro/news/same-net-same-seed-100-vs-50-on-unseen-magnitudes.txt", "jsonld": "https://wpnews.pro/news/same-net-same-seed-100-vs-50-on-unseen-magnitudes.jsonld"}}