openfree commited on
Commit
e19e8bd
路
verified 路
1 Parent(s): 1564b11
README.md CHANGED
@@ -13,4 +13,4 @@ tags: ["news", "llm", "leaderboard", "ai-media"]
13
 
14
  Engineering news and analysis from VIDRAFT: open LLMs, recursive self-improvement, leaderboards and decision AI, with reproducible numbers.
15
 
16
- Latest: [VIDRAFT in the Korean Press, 30 October 2026: ViSQL, a Hugging Face Demo, and a Benchmark Analysis](https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html), [VIDRAFT NL2SQL Engine Scores 83.80 on BIRD Mini-Dev, Above the Current #1, and Opens a Public Demo](https://vidraft-ai-media.static.hf.space/a/nl2sql-engine-bird-minidev-83-80-2026-10-29.html), [Chitos Covers All 17 Items of Korea's Financial AI Agent Security Criteria, Scores 77.8 on OWASP Benchmark On-Prem](https://vidraft-ai-media.static.hf.space/a/chitos-fsi-17-owasp-77-8-2026-10-29.html), [Decision AI, Explained: Why Hugging Face Built a Demo for Darwin-35B-ZTC-Vision](https://vidraft-ai-media.static.hf.space/a/decision-ai-ztc-vision-records-2026-10-29.html), [VIDRAFT Takes #1 on BRIGHT, the New Reasoning-Retrieval Leaderboard on Hugging Face](https://vidraft-ai-media.static.hf.space/a/bright-reasoning-retrieval-first-place-2026-10-28.html)
 
13
 
14
  Engineering news and analysis from VIDRAFT: open LLMs, recursive self-improvement, leaderboards and decision AI, with reproducible numbers.
15
 
16
+ Latest: [VIDRAFT in the Korean Press, 30 October 2026: ViSQL, NPU Pretraining, Chitos, a Hugging Face Demo and a Benchmark Analysis](https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html), [VIDRAFT NL2SQL Engine Scores 83.80 on BIRD Mini-Dev, Above the Current #1, and Opens a Public Demo](https://vidraft-ai-media.static.hf.space/a/nl2sql-engine-bird-minidev-83-80-2026-10-29.html), [Chitos Covers All 17 Items of Korea's Financial AI Agent Security Criteria, Scores 77.8 on OWASP Benchmark On-Prem](https://vidraft-ai-media.static.hf.space/a/chitos-fsi-17-owasp-77-8-2026-10-29.html), [Decision AI, Explained: Why Hugging Face Built a Demo for Darwin-35B-ZTC-Vision](https://vidraft-ai-media.static.hf.space/a/decision-ai-ztc-vision-records-2026-10-29.html), [VIDRAFT Takes #1 on BRIGHT, the New Reasoning-Retrieval Leaderboard on Hugging Face](https://vidraft-ai-media.static.hf.space/a/bright-reasoning-retrieval-first-place-2026-10-28.html)
a/vidraft-korean-press-roundup-2026-10-30.html CHANGED
@@ -1,9 +1,9 @@
1
  <!doctype html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">
2
- <title>VIDRAFT in the Korean Press, 30 October 2026: ViSQL, a Hugging Face Demo, and a Benchmark Analysis | VIDRAFT AI Media</title><meta name="description" content="Three Korean outlets covered VIDRAFT on 30 October: Chosun Biz on the ViSQL NL2SQL engine, AI Times on the Hugging Face demo, and Economist on QuantID&#x27;s benchmark analysis."><link rel="canonical" href="https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html">
3
- <meta property="og:type" content="article"><meta property="og:title" content="VIDRAFT in the Korean Press, 30 October 2026: ViSQL, a Hugging Face Demo, and a Benchmark Analysis | VIDRAFT AI Media"><meta property="og:description" content="Three Korean outlets covered VIDRAFT on 30 October: Chosun Biz on the ViSQL NL2SQL engine, AI Times on the Hugging Face demo, and Economist on QuantID&#x27;s benchmark analysis."><meta property="og:url" content="https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html"><meta property="og:image" content="https://huggingface.co/datasets/VIDraft/press-assets/resolve/main/press_20261029/2_capture_hf_demo.png">
4
  <meta name="twitter:card" content="summary_large_image"><link rel="alternate" type="application/rss+xml" title="VIDRAFT AI Media" href="https://vidraft-ai-media.static.hf.space/rss.xml">
5
  <link rel="preconnect" href="https://fonts.googleapis.com"><link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;600;800&family=Playfair+Display:wght@700&family=Space+Grotesk:wght@800&display=swap" rel="stylesheet">
6
- <script type="application/ld+json">[{"@context": "https://schema.org", "@type": "NewsArticle", "headline": "VIDRAFT in the Korean Press, 30 October 2026: ViSQL, a Hugging Face Demo, and a Benchmark Analysis", "description": "Three Korean outlets covered VIDRAFT on 30 October: Chosun Biz on the ViSQL NL2SQL engine, AI Times on the Hugging Face demo, and Economist on QuantID's benchmark analysis.", "datePublished": "2026-10-30T12:00:00+09:00", "dateModified": "2026-10-30T12:00:00+09:00", "author": {"@type": "Organization", "name": "VIDRAFT AI Media"}, "publisher": {"@type": "Organization", "name": "VIDRAFT", "url": "https://vidraft.net"}, "mainEntityOfPage": "https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html", "image": "https://huggingface.co/datasets/VIDraft/press-assets/resolve/main/press_20261029/2_capture_hf_demo.png", "keywords": "ai, news, korea, huggingface", "inLanguage": "en"}]</script><style>
7
  :root{--bg:#0a0d14;--fg:#eef1f8;--mut:#8d97ab;--ac:#5b8cff;--ac2:#9b6bff;--card:#121826;--bd:#1f2738}
8
  :root[data-theme=light]{--bg:#f7f8fc;--fg:#0b0f1a;--mut:#56607a;--card:#fff;--bd:#dfe4f0}
9
  *{box-sizing:border-box}body{margin:0;background:var(--bg);color:var(--fg);font:17px/1.75 Inter,system-ui,sans-serif}
@@ -27,8 +27,8 @@ table{border-collapse:collapse;display:block;overflow-x:auto}td,th{border:1px so
27
  blockquote{border-left:3px solid var(--ac2);margin:0;padding-left:16px;color:var(--mut)}
28
  footer{border-top:1px solid var(--bd);margin-top:60px;padding:28px 0;color:var(--mut);font-size:14px}
29
  @media(max-width:760px){.lead,.grid{grid-template-columns:1fr}.nav{display:none}}</style></head><body>
30
- <header><div class="top"><div class="w"><a class="logo" href="https://vidraft-ai-media.static.hf.space/">VIDRAFT <b>AI Media</b></a><nav class="nav"><a href="https://vidraft-ai-media.static.hf.space/">Latest</a><a href="https://vidraft-ai-media.static.hf.space/rss.xml">RSS</a><a href="https://vidraft.net">vidraft.net</a></nav></div></div></header><main><div class="rd"><article><h1>VIDRAFT in the Korean Press, 30 October 2026: ViSQL, a Hugging Face Demo, and a Benchmark Analysis</h1><p class="meta">2026-10-30 路 VIDRAFT AI Media</p><p><span class="tag">ai</span><span class="tag">news</span><span class="tag">korea</span><span class="tag">huggingface</span></p><img src="https://huggingface.co/datasets/VIDraft/press-assets/resolve/main/press_20261029/2_capture_hf_demo.png" alt="VIDRAFT in the Korean Press, 30 October 2026: ViSQL, a Hugging Face Demo, and a Benchmark Analysis"><h2 id="tldr">TL;DR</h2>
31
- <p>On 30 October 2026, three Korean outlets reported on VIDRAFT's recent work: <strong>Chosun Biz</strong> on the ViSQL text-to-SQL engine, <strong>AI Times</strong> on the demo that Hugging Face's open-source team built for our decision AI, and <strong>Economist (Korea)</strong> on QuantID's analysis of the official Hugging Face benchmark leaderboards.</p>
32
  <table>
33
  <thead>
34
  <tr>
@@ -53,6 +53,16 @@ footer{border-top:1px solid var(--bd);margin-top:60px;padding:28px 0;color:var(-
53
  <td>QuantID: 53 official Hugging Face benchmarks, Korea second in #1 spots</td>
54
  <td><a href="https://economist.co.kr/article/view/ecn202610300010">Read</a></td>
55
  </tr>
 
 
 
 
 
 
 
 
 
 
56
  </tbody>
57
  </table>
58
  <h2 id="visql-chosun-biz">ViSQL (Chosun Biz)</h2>
@@ -70,5 +80,13 @@ footer{border-top:1px solid var(--bd);margin-top:60px;padding:28px 0;color:var(-
70
  <p>QuantID counted 53 official Hugging Face benchmarks with 48 leaderboards on 29 October. By the publisher's country, China holds 24 #1 spots, Korea 15 and the USA 2. All 15 Korean #1 spots are FINAL-Bench (VIDRAFT) entries, scored as publisher-reported on Hugging Face. QuantID is a technology partner of VIDRAFT.</p>
71
  <ul>
72
  <li>Map: <a href="https://huggingface.co/spaces/quantid/huggingface-official-benchmark-leaderboards">Hugging Face Official Benchmarks Map</a></li>
 
 
 
 
 
 
 
 
73
  </ul></article></div></main>
74
  <footer><div class="w">漏 VIDRAFT 路 <a href="https://vidraft.net">vidraft.net</a> 路 <a href="https://vidraft-ai-media.static.hf.space/rss.xml">RSS</a> 路 <a href="https://vidraft-ai-media.static.hf.space/sitemap.xml">Sitemap</a></div></footer></body></html>
 
1
  <!doctype html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">
2
+ <title>VIDRAFT in the Korean Press, 30 October 2026: ViSQL, NPU Pretraining, Chitos, a Hugging Face Demo and a Benchmark Analysis | VIDRAFT AI Media</title><meta name="description" content="Five Korean outlets covered VIDRAFT on 30 October: Chosun Biz on ViSQL, AI Times on the Hugging Face demo, Economist on QuantID&#x27;s benchmark analysis, Financial News on Chitos, and JoongAng Ilbo on NPU pretraining."><link rel="canonical" href="https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html">
3
+ <meta property="og:type" content="article"><meta property="og:title" content="VIDRAFT in the Korean Press, 30 October 2026: ViSQL, NPU Pretraining, Chitos, a Hugging Face Demo and a Benchmark Analysis | VIDRAFT AI Media"><meta property="og:description" content="Five Korean outlets covered VIDRAFT on 30 October: Chosun Biz on ViSQL, AI Times on the Hugging Face demo, Economist on QuantID&#x27;s benchmark analysis, Financial News on Chitos, and JoongAng Ilbo on NPU pretraining."><meta property="og:url" content="https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html"><meta property="og:image" content="https://huggingface.co/datasets/VIDraft/press-assets/resolve/main/press_20261029/2_capture_hf_demo.png">
4
  <meta name="twitter:card" content="summary_large_image"><link rel="alternate" type="application/rss+xml" title="VIDRAFT AI Media" href="https://vidraft-ai-media.static.hf.space/rss.xml">
5
  <link rel="preconnect" href="https://fonts.googleapis.com"><link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;600;800&family=Playfair+Display:wght@700&family=Space+Grotesk:wght@800&display=swap" rel="stylesheet">
6
+ <script type="application/ld+json">[{"@context": "https://schema.org", "@type": "NewsArticle", "headline": "VIDRAFT in the Korean Press, 30 October 2026: ViSQL, NPU Pretraining, Chitos, a Hugging Face Demo and a Benchm", "description": "Five Korean outlets covered VIDRAFT on 30 October: Chosun Biz on ViSQL, AI Times on the Hugging Face demo, Economist on QuantID's benchmark analysis, Financial News on Chitos, and JoongAng Ilbo on NPU pretraining.", "datePublished": "2026-10-30T12:00:00+09:00", "dateModified": "2026-10-30T12:00:00+09:00", "author": {"@type": "Organization", "name": "VIDRAFT AI Media"}, "publisher": {"@type": "Organization", "name": "VIDRAFT", "url": "https://vidraft.net"}, "mainEntityOfPage": "https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html", "image": "https://huggingface.co/datasets/VIDraft/press-assets/resolve/main/press_20261029/2_capture_hf_demo.png", "keywords": "ai, news, korea, huggingface", "inLanguage": "en"}]</script><style>
7
  :root{--bg:#0a0d14;--fg:#eef1f8;--mut:#8d97ab;--ac:#5b8cff;--ac2:#9b6bff;--card:#121826;--bd:#1f2738}
8
  :root[data-theme=light]{--bg:#f7f8fc;--fg:#0b0f1a;--mut:#56607a;--card:#fff;--bd:#dfe4f0}
9
  *{box-sizing:border-box}body{margin:0;background:var(--bg);color:var(--fg);font:17px/1.75 Inter,system-ui,sans-serif}
 
27
  blockquote{border-left:3px solid var(--ac2);margin:0;padding-left:16px;color:var(--mut)}
28
  footer{border-top:1px solid var(--bd);margin-top:60px;padding:28px 0;color:var(--mut);font-size:14px}
29
  @media(max-width:760px){.lead,.grid{grid-template-columns:1fr}.nav{display:none}}</style></head><body>
30
+ <header><div class="top"><div class="w"><a class="logo" href="https://vidraft-ai-media.static.hf.space/">VIDRAFT <b>AI Media</b></a><nav class="nav"><a href="https://vidraft-ai-media.static.hf.space/">Latest</a><a href="https://vidraft-ai-media.static.hf.space/rss.xml">RSS</a><a href="https://vidraft.net">vidraft.net</a></nav></div></div></header><main><div class="rd"><article><h1>VIDRAFT in the Korean Press, 30 October 2026: ViSQL, NPU Pretraining, Chitos, a Hugging Face Demo and a Benchmark Analysis</h1><p class="meta">2026-10-30 路 VIDRAFT AI Media</p><p><span class="tag">ai</span><span class="tag">news</span><span class="tag">korea</span><span class="tag">huggingface</span></p><img src="https://huggingface.co/datasets/VIDraft/press-assets/resolve/main/press_20261029/2_capture_hf_demo.png" alt="VIDRAFT in the Korean Press, 30 October 2026: ViSQL, NPU Pretraining, Chitos, a Hugging Face Demo and a Benchmark Analysis"><h2 id="tldr">TL;DR</h2>
31
+ <p>On 30 October 2026, five Korean outlets reported on VIDRAFT's recent work: <strong>Chosun Biz</strong> on the ViSQL text-to-SQL engine, <strong>AI Times</strong> on the demo that Hugging Face's open-source team built for our decision AI, and <strong>Economist (Korea)</strong> on QuantID's analysis of the official Hugging Face benchmark leaderboards. <strong>Financial News</strong> covered the Chitos security engine, and <strong>JoongAng Ilbo</strong> covered pretraining a language model on a Korean inference NPU.</p>
32
  <table>
33
  <thead>
34
  <tr>
 
53
  <td>QuantID: 53 official Hugging Face benchmarks, Korea second in #1 spots</td>
54
  <td><a href="https://economist.co.kr/article/view/ecn202610300010">Read</a></td>
55
  </tr>
56
+ <tr>
57
+ <td>Financial News</td>
58
+ <td>Chitos: modules covering all 17 FSI AI agent security evaluation criteria</td>
59
+ <td><a href="https://www.fnnews.com/news/202610301324484711">Read</a></td>
60
+ </tr>
61
+ <tr>
62
+ <td>JoongAng Ilbo</td>
63
+ <td>Pretraining a language model from scratch on a Korean inference NPU</td>
64
+ <td><a href="https://www.joongang.co.kr/article/25474466">Read</a></td>
65
+ </tr>
66
  </tbody>
67
  </table>
68
  <h2 id="visql-chosun-biz">ViSQL (Chosun Biz)</h2>
 
80
  <p>QuantID counted 53 official Hugging Face benchmarks with 48 leaderboards on 29 October. By the publisher's country, China holds 24 #1 spots, Korea 15 and the USA 2. All 15 Korean #1 spots are FINAL-Bench (VIDRAFT) entries, scored as publisher-reported on Hugging Face. QuantID is a technology partner of VIDRAFT.</p>
81
  <ul>
82
  <li>Map: <a href="https://huggingface.co/spaces/quantid/huggingface-official-benchmark-leaderboards">Hugging Face Official Benchmarks Map</a></li>
83
+ </ul>
84
+ <h2 id="chitos-security-engine-financial-news">Chitos security engine (Financial News)</h2>
85
+ <p>Chitos, our AI agent security engine, now has modules that address all 17 evaluation criteria in the Financial Security Institute's AI agent security framework. It also runs fully on premises, for networks with no internet access.</p>
86
+ <h2 id="npu-pretraining-joongang-ilbo">NPU pretraining (JoongAng Ilbo)</h2>
87
+ <p>JoongAng Ilbo reported that VIDRAFT pretrained <strong>VKORE-325M-NPU-Base</strong> (325.6M parameters) from random initialization on Rebellions ATOM-Max, a Korean NPU that was released for inference. To compare fairly, we trained the same model on an NVIDIA B200 GPU with the same initial weights, data, batch size and learning-rate schedule: 76,292 steps on FineWeb-Edu, about four days, with no divergence. The weights and evaluation scripts of both models are public.</p>
88
+ <p>Important: this works only when VIDRAFT's own kernel is installed on the NPU. The training method is patent pending.</p>
89
+ <ul>
90
+ <li>Model: <a href="https://huggingface.co/FINAL-Bench/VKORE-325M-NPU-Base">FINAL-Bench/VKORE-325M-NPU-Base</a></li>
91
  </ul></article></div></main>
92
  <footer><div class="w">漏 VIDRAFT 路 <a href="https://vidraft.net">vidraft.net</a> 路 <a href="https://vidraft-ai-media.static.hf.space/rss.xml">RSS</a> 路 <a href="https://vidraft-ai-media.static.hf.space/sitemap.xml">Sitemap</a></div></footer></body></html>
index.html CHANGED
@@ -27,5 +27,5 @@ table{border-collapse:collapse;display:block;overflow-x:auto}td,th{border:1px so
27
  blockquote{border-left:3px solid var(--ac2);margin:0;padding-left:16px;color:var(--mut)}
28
  footer{border-top:1px solid var(--bd);margin-top:60px;padding:28px 0;color:var(--mut);font-size:14px}
29
  @media(max-width:760px){.lead,.grid{grid-template-columns:1fr}.nav{display:none}}</style></head><body>
30
- <header><div class="top"><div class="w"><a class="logo" href="https://vidraft-ai-media.static.hf.space/">VIDRAFT <b>AI Media</b></a><nav class="nav"><a href="https://vidraft-ai-media.static.hf.space/">Latest</a><a href="https://vidraft-ai-media.static.hf.space/rss.xml">RSS</a><a href="https://vidraft.net">vidraft.net</a></nav></div></div></header><main><div class="w"><section class="hero"><div class="kick">Engineering newsroom</div><h1>VIDRAFT AI Media</h1><p>Engineering news and analysis from VIDRAFT: open LLMs, recursive self-improvement, leaderboards and decision AI, with reproducible numbers.</p></section><a class="lead" href="https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html"><div><img src="https://huggingface.co/datasets/VIDraft/press-assets/resolve/main/press_20261029/2_capture_hf_demo.png" alt=""></div><div><span class="tag">Lead story</span><h2>VIDRAFT in the Korean Press, 30 October 2026: ViSQL, a Hugging Face Demo, and a Benchmark Analysis</h2><p class="meta">2026-10-30</p><p>Three Korean outlets covered VIDRAFT on 30 October: Chosun Biz on the ViSQL NL2SQL engine, AI Times on the Hugging Face demo, and Economist on QuantID&#x27;s benchmark analysis.</p></div></a><div class="grid"><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/nl2sql-engine-bird-minidev-83-80-2026-10-29.html"><span class=tag>ai</span><span class=tag>sql</span><h3>VIDRAFT NL2SQL Engine Scores 83.80 on BIRD Mini-Dev, Above the Current #1, and Opens a Public Demo</h3><p class="meta">2026-10-29</p><p>Darwin-180B-NL2SQL reaches 83.80% execution accuracy on BIRD Mini-Dev (SQLite), above the listed #1 at 81.00. A public demo drafts a legal fraud notice from data.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/chitos-fsi-17-owasp-77-8-2026-10-29.html"><span class=tag>security</span><span class=tag>ai</span><h3>Chitos Covers All 17 Items of Korea&#x27;s Financial AI Agent Security Criteria, Scores 77.8 on OWASP Benchmark On-Prem</h3><p class="meta">2026-10-29</p><p>VIDRAFT&#x27;s Chitos has modules for all 17 items of Korea&#x27;s financial AI agent security criteria and scores 77.8 (Youden) on OWASP Benchmark with an on-premises engine.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/decision-ai-ztc-vision-records-2026-10-29.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>Decision AI, Explained: Why Hugging Face Built a Demo for Darwin-35B-ZTC-Vision</h3><p class="meta">2026-10-29</p><p>Darwin-35B-ZTC-Vision answers with a calibrated probability per option in one pass. Hugging Face&#x27;s open-source team built and donated its demo. Records inside.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/bright-reasoning-retrieval-first-place-2026-10-28.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>VIDRAFT Takes #1 on BRIGHT, the New Reasoning-Retrieval Leaderboard on Hugging Face</h3><p class="meta">2026-10-28</p><p>FINAL-Bench scores 0.307 nDCG@10 on BRIGHT, Hugging Face&#x27;s new reasoning-intensive retrieval leaderboard. VIDRAFT now leads 12 of 52 official benchmarks.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/darwin-35b-ztc-vision-hugging-face-demo-2026-10-28.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>Hugging Face&#x27;s Open-Source Team Built a Free Demo for Darwin-35B-ZTC-Vision</h3><p class="meta">2026-10-28</p><p>Hugging Face&#x27;s open-source team built a free ZeroGPU demo of Darwin-35B-ZTC-Vision, VIDRAFT&#x27;s 35B vision decision model, and is moving it to FINAL-Bench.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/freeweb-security-scan-free-website-check-2026-10-27.html"><span class=tag>security</span><span class=tag>webdev</span><h3>Free 1-Minute Website Check for the Age of AI Hacking: Security, SEO and AI Visibility</h3><p class="meta">2026-10-27</p><p>FreeWeb Security SCAN gives a read-only report on security headers, HTTPS, email spoofing protection, SEO basics and access for 9 AI crawlers.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/vkore-325m-npu-pretraining-atom-max-2026-10-26.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>VKORE-325M: A 325M LLM Pretrained From Zero on a Rebellions ATOM-Max NPU, Equal to a B200 Twin</h3><p class="meta">2026-10-26</p><p>VIDRAFT pretrained a 325M model from random init on an inference-only Rebellions ATOM-Max NPU. On unseen text: 0.7993 vs 0.7994 for the same model on a B200.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/bf16-residual-stream-stall-long-training-2026-10-22.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>Why Long bf16 Training Runs Stall: The Residual Stream Has No Room for Small Updates</h3><p class="meta">2026-10-22</p><p>A bf16 residual near 600 has a resolution of 4, so small updates vanish. Measured gaps, a reproducible simulation and fixes for long training runs.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/darwin-27b-ztc-v9-s1mb-number-one-2026-10-20.html"><span class=tag>ai</span><span class=tag>llm</span><h3>Darwin-27B-ZTC-v9 Takes #1 on S1MB Across 197 Decision Models</h3><p class="meta">2026-10-20</p><p>A 27B decision model that answers without generating text now leads the S1MB English leaderboard on Borda score across 137 benchmarks, ahead of entries from Perplexity, Meta and Cloudflare.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/ztc-decision-model-one-forward-pass-2026-10-19.html"><span class=tag>ai</span><span class=tag>llm</span><h3>ZTC: A Decision Model That Answers in One Forward Pass</h3><p class="meta">2026-10-19</p><p>Darwin-27B-ZTC-v5 returns a calibrated probability for every option without generating text. How decision models work, how they are scored, and how to serve one with transformers serve.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/leadboard-held-out-labels-2026-10-19.html"><span class=tag>ai</span><span class=tag>drugdiscovery</span><h3>LEADBOARD: A Drug Property Benchmark Whose Labels Never Leave the Verifier</h3><p class="meta">2026-10-19</p><p>21 boards across 7 disciplines and 18,382 held-out compounds. Time splits, published noise floors and baselines, and a ladder rule so the board cannot be won by resubmitting.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-discovery-challenge-four-seasons-2026-10-19.html"><span class=tag>ai</span><span class=tag>drugdiscovery</span><h3>Open Discovery Challenge: Four Diseases, Four Targets, One Verifier</h3><p class="meta">2026-10-19</p><p>An open drug discovery competition. Submit a SMILES string; a hidden pipeline scores binding, selectivity, ADMET, novelty and synthesis against a target and a counter target.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-materials-challenge-solid-state-electrolyte-2026-10-19.html"><span class=tag>ai</span><span class=tag>materials</span><h3>Open Materials Challenge: Find a Solid-State Battery Electrolyte, Get It Verified</h3><p class="meta">2026-10-19</p><p>Submit a lithium-containing crystal structure. A hidden pipeline checks oxidation stability, lithium-metal stability and novelty of use. Season 1 closes 2026-11-30.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge-gpu-only-2026-10-19.html"><span class=tag>ai</span><span class=tag>quantum</span><h3>Open Quantum Challenge: Two Quantum Seasons, No Quantum Hardware Required</h3><p class="meta">2026-10-19</p><p>Compute Fermi-Hubbard ground-state energies or build a surface-code decoder with a GPU and an open harness. Answers are held privately and every entry is scored automatically.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-science-challenges-verifier-decides-2026-10-19.html"><span class=tag>ai</span><span class=tag>science</span><h3>Five Open Science Challenges Where a Verifier, Not the Submitter, Decides the Score</h3><p class="meta">2026-10-19</p><p>Drug discovery, battery materials, superconductors, quantum computing and drug property prediction: five open challenges from FINAL-Bench, each graded by a verifier that holds the answer privately.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-superconductor-challenge-laptop-2026-10-19.html"><span class=tag>ai</span><span class=tag>physics</span><h3>Open Superconductor Challenge: Screen 2D Materials for d-Wave Pairing on a Laptop</h3><p class="meta">2026-10-19</p><p>Estimate d-wave pairing tendency from a published effective Hubbard model per material. A many-body reference solver verifies the top entries. Season 1 closes 2026-12-31.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/vidraft-open-source-github-six-projects.html"><span class=tag>opensource</span><span class=tag>ai</span><h3>Six Open-Source Projects from VIDRAFT, Now on GitHub</h3><p class="meta">2026-10-18</p><p>ONGRID, FINAL Bench, the Open Quantum Challenge harness, POCKET, ZTC and ALL Bench Leaderboard: what each project does and where to start.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/vidraft-week-2026-10-17.html"><span class=tag>ai</span><span class=tag>llm</span><h3>VIDRAFT This Week: Typed Decisions #1, S1MB Japanese #1, and NPU Pretraining in the Press</h3><p class="meta">2026-10-17</p><p>A weekly roundup: Darwin-27B-ZTC-v5 holds #1 on Hugging Face Typed Decisions at 0.81, the S1MB Japanese edition #1 at 84.57, and national press coverage of VKORE pretraining on inference-only NPUs.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/typed-decisions-first-eleven-benchmarks.html"><span class=tag>ai</span><span class=tag>llm</span><h3>VIDRAFT Now Leads 11 Hugging Face Official Benchmarks After Reclaiming #1 on Typed Decisions</h3><p class="meta">2026-10-16</p><p>Darwin-27B-ZTC-v5 reclaims #1 on the Typed Decisions official benchmark at 0.81, holding both first and second place, bringing VIDRAFT to #1 on 11 Hugging Face official benchmarks.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/vkore-46m-npu-pretraining-inference-only.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>Pretraining From Scratch on an Inference-Only NPU: VKORE-46M-NPU</h3><p class="meta">2026-10-14</p><p>VIDRAFT pretrained a language model from scratch on an inference-only NPU via real backpropagation and released it on Hugging Face. Loss trajectory correlates 0.99 with a GPU reference.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/model-level-rsi-what-it-means.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>Model-Level RSI: What It Means When a Model Improves Its Own Weights</h3><p class="meta">2026-10-12</p><p>A practical look at model-level recursive self-improvement, the five stages, and how to read verifiable results like 100% on AIME 2026 and 93.43% on GPQA Diamond.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge.html"><span class=tag>ai</span><span class=tag>quantum</span><h3>Open Quantum Challenge: Solve Quantum Problems With a GPU and No Quantum Computer</h3><p class="meta">2026-10-11</p><p>The Open Quantum Challenge on Hugging Face opens with two seasons, quantum simulation and QEC decoding. Use only a GPU and an open harness, no quantum hardware. 2,000 USD in prizes.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/ai-native-vidraft.html"><span class=tag>ai</span><span class=tag>agi</span><h3>AI Native: Ten People, Hundreds of AI, and the Arrival of the Pre-AGI Age</h3><p class="meta">2026-10-11</p><p>VIDRAFT is an AI-native lab where hundreds of AI instances run as a company on a few dozen GPUs. Perfect math scores, 165M downloads, quantum hardware, and a self-improving org point to the Pre-AGI age.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/s1mb-number-one-decision-engine.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>S1MB Number One: How a Zero-Token Judge Won the Decision-Engine Leaderboard</h3><p class="meta">2026-10-11</p><p>Our model ranked number one on the S1MB decision-engine leaderboard among 102 models, with a Borda score of 89.58, deciding in a single forward pass with zero generated tokens.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/typed-decision-ai-a16z-vs-open-ztc.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>a16z Bet $870M on Typed, Machine-Native AI. Here Is the Open, Benchmark-Proven Version</h3><p class="meta">2026-10-10</p><p>TypeSafe AI raised 870M dollars for typed, machine-native AI. We explain what typed decision AI is, why it matters, and show an open, zero-token, S1MB number one alternative.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-180b-model-leads-nine-official-hugging-face-leaderboards.html"><span class=tag>ai</span><span class=tag>llm</span><h3>From exams to real work: an open 180B model now leads nine official Hugging Face leaderboards</h3><p class="meta">2026-10-05</p><p>Darwin-180B-RSI adds #1 on IFStruct (structured output, 98.95%) and ExtractBench (document extraction, 90.29), for nine first places among 48 official Hugg</p></a></div></div></main>
31
  <footer><div class="w">漏 VIDRAFT 路 <a href="https://vidraft.net">vidraft.net</a> 路 <a href="https://vidraft-ai-media.static.hf.space/rss.xml">RSS</a> 路 <a href="https://vidraft-ai-media.static.hf.space/sitemap.xml">Sitemap</a></div></footer></body></html>
 
27
  blockquote{border-left:3px solid var(--ac2);margin:0;padding-left:16px;color:var(--mut)}
28
  footer{border-top:1px solid var(--bd);margin-top:60px;padding:28px 0;color:var(--mut);font-size:14px}
29
  @media(max-width:760px){.lead,.grid{grid-template-columns:1fr}.nav{display:none}}</style></head><body>
30
+ <header><div class="top"><div class="w"><a class="logo" href="https://vidraft-ai-media.static.hf.space/">VIDRAFT <b>AI Media</b></a><nav class="nav"><a href="https://vidraft-ai-media.static.hf.space/">Latest</a><a href="https://vidraft-ai-media.static.hf.space/rss.xml">RSS</a><a href="https://vidraft.net">vidraft.net</a></nav></div></div></header><main><div class="w"><section class="hero"><div class="kick">Engineering newsroom</div><h1>VIDRAFT AI Media</h1><p>Engineering news and analysis from VIDRAFT: open LLMs, recursive self-improvement, leaderboards and decision AI, with reproducible numbers.</p></section><a class="lead" href="https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html"><div><img src="https://huggingface.co/datasets/VIDraft/press-assets/resolve/main/press_20261029/2_capture_hf_demo.png" alt=""></div><div><span class="tag">Lead story</span><h2>VIDRAFT in the Korean Press, 30 October 2026: ViSQL, NPU Pretraining, Chitos, a Hugging Face Demo and a Benchmark Analysis</h2><p class="meta">2026-10-30</p><p>Five Korean outlets covered VIDRAFT on 30 October: Chosun Biz on ViSQL, AI Times on the Hugging Face demo, Economist on QuantID&#x27;s benchmark analysis, Financial News on Chitos, and JoongAng Ilbo on NPU pretraining.</p></div></a><div class="grid"><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/nl2sql-engine-bird-minidev-83-80-2026-10-29.html"><span class=tag>ai</span><span class=tag>sql</span><h3>VIDRAFT NL2SQL Engine Scores 83.80 on BIRD Mini-Dev, Above the Current #1, and Opens a Public Demo</h3><p class="meta">2026-10-29</p><p>Darwin-180B-NL2SQL reaches 83.80% execution accuracy on BIRD Mini-Dev (SQLite), above the listed #1 at 81.00. A public demo drafts a legal fraud notice from data.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/chitos-fsi-17-owasp-77-8-2026-10-29.html"><span class=tag>security</span><span class=tag>ai</span><h3>Chitos Covers All 17 Items of Korea&#x27;s Financial AI Agent Security Criteria, Scores 77.8 on OWASP Benchmark On-Prem</h3><p class="meta">2026-10-29</p><p>VIDRAFT&#x27;s Chitos has modules for all 17 items of Korea&#x27;s financial AI agent security criteria and scores 77.8 (Youden) on OWASP Benchmark with an on-premises engine.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/decision-ai-ztc-vision-records-2026-10-29.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>Decision AI, Explained: Why Hugging Face Built a Demo for Darwin-35B-ZTC-Vision</h3><p class="meta">2026-10-29</p><p>Darwin-35B-ZTC-Vision answers with a calibrated probability per option in one pass. Hugging Face&#x27;s open-source team built and donated its demo. Records inside.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/bright-reasoning-retrieval-first-place-2026-10-28.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>VIDRAFT Takes #1 on BRIGHT, the New Reasoning-Retrieval Leaderboard on Hugging Face</h3><p class="meta">2026-10-28</p><p>FINAL-Bench scores 0.307 nDCG@10 on BRIGHT, Hugging Face&#x27;s new reasoning-intensive retrieval leaderboard. VIDRAFT now leads 12 of 52 official benchmarks.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/darwin-35b-ztc-vision-hugging-face-demo-2026-10-28.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>Hugging Face&#x27;s Open-Source Team Built a Free Demo for Darwin-35B-ZTC-Vision</h3><p class="meta">2026-10-28</p><p>Hugging Face&#x27;s open-source team built a free ZeroGPU demo of Darwin-35B-ZTC-Vision, VIDRAFT&#x27;s 35B vision decision model, and is moving it to FINAL-Bench.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/freeweb-security-scan-free-website-check-2026-10-27.html"><span class=tag>security</span><span class=tag>webdev</span><h3>Free 1-Minute Website Check for the Age of AI Hacking: Security, SEO and AI Visibility</h3><p class="meta">2026-10-27</p><p>FreeWeb Security SCAN gives a read-only report on security headers, HTTPS, email spoofing protection, SEO basics and access for 9 AI crawlers.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/vkore-325m-npu-pretraining-atom-max-2026-10-26.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>VKORE-325M: A 325M LLM Pretrained From Zero on a Rebellions ATOM-Max NPU, Equal to a B200 Twin</h3><p class="meta">2026-10-26</p><p>VIDRAFT pretrained a 325M model from random init on an inference-only Rebellions ATOM-Max NPU. On unseen text: 0.7993 vs 0.7994 for the same model on a B200.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/bf16-residual-stream-stall-long-training-2026-10-22.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>Why Long bf16 Training Runs Stall: The Residual Stream Has No Room for Small Updates</h3><p class="meta">2026-10-22</p><p>A bf16 residual near 600 has a resolution of 4, so small updates vanish. Measured gaps, a reproducible simulation and fixes for long training runs.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/darwin-27b-ztc-v9-s1mb-number-one-2026-10-20.html"><span class=tag>ai</span><span class=tag>llm</span><h3>Darwin-27B-ZTC-v9 Takes #1 on S1MB Across 197 Decision Models</h3><p class="meta">2026-10-20</p><p>A 27B decision model that answers without generating text now leads the S1MB English leaderboard on Borda score across 137 benchmarks, ahead of entries from Perplexity, Meta and Cloudflare.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/ztc-decision-model-one-forward-pass-2026-10-19.html"><span class=tag>ai</span><span class=tag>llm</span><h3>ZTC: A Decision Model That Answers in One Forward Pass</h3><p class="meta">2026-10-19</p><p>Darwin-27B-ZTC-v5 returns a calibrated probability for every option without generating text. How decision models work, how they are scored, and how to serve one with transformers serve.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/leadboard-held-out-labels-2026-10-19.html"><span class=tag>ai</span><span class=tag>drugdiscovery</span><h3>LEADBOARD: A Drug Property Benchmark Whose Labels Never Leave the Verifier</h3><p class="meta">2026-10-19</p><p>21 boards across 7 disciplines and 18,382 held-out compounds. Time splits, published noise floors and baselines, and a ladder rule so the board cannot be won by resubmitting.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-discovery-challenge-four-seasons-2026-10-19.html"><span class=tag>ai</span><span class=tag>drugdiscovery</span><h3>Open Discovery Challenge: Four Diseases, Four Targets, One Verifier</h3><p class="meta">2026-10-19</p><p>An open drug discovery competition. Submit a SMILES string; a hidden pipeline scores binding, selectivity, ADMET, novelty and synthesis against a target and a counter target.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-materials-challenge-solid-state-electrolyte-2026-10-19.html"><span class=tag>ai</span><span class=tag>materials</span><h3>Open Materials Challenge: Find a Solid-State Battery Electrolyte, Get It Verified</h3><p class="meta">2026-10-19</p><p>Submit a lithium-containing crystal structure. A hidden pipeline checks oxidation stability, lithium-metal stability and novelty of use. Season 1 closes 2026-11-30.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge-gpu-only-2026-10-19.html"><span class=tag>ai</span><span class=tag>quantum</span><h3>Open Quantum Challenge: Two Quantum Seasons, No Quantum Hardware Required</h3><p class="meta">2026-10-19</p><p>Compute Fermi-Hubbard ground-state energies or build a surface-code decoder with a GPU and an open harness. Answers are held privately and every entry is scored automatically.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-science-challenges-verifier-decides-2026-10-19.html"><span class=tag>ai</span><span class=tag>science</span><h3>Five Open Science Challenges Where a Verifier, Not the Submitter, Decides the Score</h3><p class="meta">2026-10-19</p><p>Drug discovery, battery materials, superconductors, quantum computing and drug property prediction: five open challenges from FINAL-Bench, each graded by a verifier that holds the answer privately.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-superconductor-challenge-laptop-2026-10-19.html"><span class=tag>ai</span><span class=tag>physics</span><h3>Open Superconductor Challenge: Screen 2D Materials for d-Wave Pairing on a Laptop</h3><p class="meta">2026-10-19</p><p>Estimate d-wave pairing tendency from a published effective Hubbard model per material. A many-body reference solver verifies the top entries. Season 1 closes 2026-12-31.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/vidraft-open-source-github-six-projects.html"><span class=tag>opensource</span><span class=tag>ai</span><h3>Six Open-Source Projects from VIDRAFT, Now on GitHub</h3><p class="meta">2026-10-18</p><p>ONGRID, FINAL Bench, the Open Quantum Challenge harness, POCKET, ZTC and ALL Bench Leaderboard: what each project does and where to start.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/vidraft-week-2026-10-17.html"><span class=tag>ai</span><span class=tag>llm</span><h3>VIDRAFT This Week: Typed Decisions #1, S1MB Japanese #1, and NPU Pretraining in the Press</h3><p class="meta">2026-10-17</p><p>A weekly roundup: Darwin-27B-ZTC-v5 holds #1 on Hugging Face Typed Decisions at 0.81, the S1MB Japanese edition #1 at 84.57, and national press coverage of VKORE pretraining on inference-only NPUs.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/typed-decisions-first-eleven-benchmarks.html"><span class=tag>ai</span><span class=tag>llm</span><h3>VIDRAFT Now Leads 11 Hugging Face Official Benchmarks After Reclaiming #1 on Typed Decisions</h3><p class="meta">2026-10-16</p><p>Darwin-27B-ZTC-v5 reclaims #1 on the Typed Decisions official benchmark at 0.81, holding both first and second place, bringing VIDRAFT to #1 on 11 Hugging Face official benchmarks.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/vkore-46m-npu-pretraining-inference-only.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>Pretraining From Scratch on an Inference-Only NPU: VKORE-46M-NPU</h3><p class="meta">2026-10-14</p><p>VIDRAFT pretrained a language model from scratch on an inference-only NPU via real backpropagation and released it on Hugging Face. Loss trajectory correlates 0.99 with a GPU reference.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/model-level-rsi-what-it-means.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>Model-Level RSI: What It Means When a Model Improves Its Own Weights</h3><p class="meta">2026-10-12</p><p>A practical look at model-level recursive self-improvement, the five stages, and how to read verifiable results like 100% on AIME 2026 and 93.43% on GPQA Diamond.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge.html"><span class=tag>ai</span><span class=tag>quantum</span><h3>Open Quantum Challenge: Solve Quantum Problems With a GPU and No Quantum Computer</h3><p class="meta">2026-10-11</p><p>The Open Quantum Challenge on Hugging Face opens with two seasons, quantum simulation and QEC decoding. Use only a GPU and an open harness, no quantum hardware. 2,000 USD in prizes.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/ai-native-vidraft.html"><span class=tag>ai</span><span class=tag>agi</span><h3>AI Native: Ten People, Hundreds of AI, and the Arrival of the Pre-AGI Age</h3><p class="meta">2026-10-11</p><p>VIDRAFT is an AI-native lab where hundreds of AI instances run as a company on a few dozen GPUs. Perfect math scores, 165M downloads, quantum hardware, and a self-improving org point to the Pre-AGI age.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/s1mb-number-one-decision-engine.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>S1MB Number One: How a Zero-Token Judge Won the Decision-Engine Leaderboard</h3><p class="meta">2026-10-11</p><p>Our model ranked number one on the S1MB decision-engine leaderboard among 102 models, with a Borda score of 89.58, deciding in a single forward pass with zero generated tokens.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/typed-decision-ai-a16z-vs-open-ztc.html"><span class=tag>ai</span><span class=tag>machinelearning</span><h3>a16z Bet $870M on Typed, Machine-Native AI. Here Is the Open, Benchmark-Proven Version</h3><p class="meta">2026-10-10</p><p>TypeSafe AI raised 870M dollars for typed, machine-native AI. We explain what typed decision AI is, why it matters, and show an open, zero-token, S1MB number one alternative.</p></a><a class="cell" href="https://vidraft-ai-media.static.hf.space/a/open-180b-model-leads-nine-official-hugging-face-leaderboards.html"><span class=tag>ai</span><span class=tag>llm</span><h3>From exams to real work: an open 180B model now leads nine official Hugging Face leaderboards</h3><p class="meta">2026-10-05</p><p>Darwin-180B-RSI adds #1 on IFStruct (structured output, 98.95%) and ExtractBench (document extraction, 90.29), for nine first places among 48 official Hugg</p></a></div></div></main>
31
  <footer><div class="w">漏 VIDRAFT 路 <a href="https://vidraft.net">vidraft.net</a> 路 <a href="https://vidraft-ai-media.static.hf.space/rss.xml">RSS</a> 路 <a href="https://vidraft-ai-media.static.hf.space/sitemap.xml">Sitemap</a></div></footer></body></html>
rss.xml CHANGED
@@ -1 +1 @@
1
- <?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>VIDRAFT AI Media</title><link>https://vidraft-ai-media.static.hf.space/</link><description>Engineering news and analysis from VIDRAFT: open LLMs, recursive self-improvement, leaderboards and decision AI, with reproducible numbers.</description><item><title>VIDRAFT in the Korean Press, 30 October 2026: ViSQL, a Hugging Face Demo, and a Benchmark Analysis</title><link>https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html</link><guid>https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html</guid><description>Three Korean outlets covered VIDRAFT on 30 October: Chosun Biz on the ViSQL NL2SQL engine, AI Times on the Hugging Face demo, and Economist on QuantID&#x27;s benchmark analysis.</description><pubDate>2026-10-30T12:00:00+09:00</pubDate></item><item><title>VIDRAFT NL2SQL Engine Scores 83.80 on BIRD Mini-Dev, Above the Current #1, and Opens a Public Demo</title><link>https://vidraft-ai-media.static.hf.space/a/nl2sql-engine-bird-minidev-83-80-2026-10-29.html</link><guid>https://vidraft-ai-media.static.hf.space/a/nl2sql-engine-bird-minidev-83-80-2026-10-29.html</guid><description>Darwin-180B-NL2SQL reaches 83.80% execution accuracy on BIRD Mini-Dev (SQLite), above the listed #1 at 81.00. A public demo drafts a legal fraud notice from data.</description><pubDate>2026-10-29T18:00:00+09:00</pubDate></item><item><title>Chitos Covers All 17 Items of Korea&#x27;s Financial AI Agent Security Criteria, Scores 77.8 on OWASP Benchmark On-Prem</title><link>https://vidraft-ai-media.static.hf.space/a/chitos-fsi-17-owasp-77-8-2026-10-29.html</link><guid>https://vidraft-ai-media.static.hf.space/a/chitos-fsi-17-owasp-77-8-2026-10-29.html</guid><description>VIDRAFT&#x27;s Chitos has modules for all 17 items of Korea&#x27;s financial AI agent security criteria and scores 77.8 (Youden) on OWASP Benchmark with an on-premises engine.</description><pubDate>2026-10-29T17:40:00+09:00</pubDate></item><item><title>Decision AI, Explained: Why Hugging Face Built a Demo for Darwin-35B-ZTC-Vision</title><link>https://vidraft-ai-media.static.hf.space/a/decision-ai-ztc-vision-records-2026-10-29.html</link><guid>https://vidraft-ai-media.static.hf.space/a/decision-ai-ztc-vision-records-2026-10-29.html</guid><description>Darwin-35B-ZTC-Vision answers with a calibrated probability per option in one pass. Hugging Face&#x27;s open-source team built and donated its demo. Records inside.</description><pubDate>2026-10-29T17:20:00+09:00</pubDate></item><item><title>VIDRAFT Takes #1 on BRIGHT, the New Reasoning-Retrieval Leaderboard on Hugging Face</title><link>https://vidraft-ai-media.static.hf.space/a/bright-reasoning-retrieval-first-place-2026-10-28.html</link><guid>https://vidraft-ai-media.static.hf.space/a/bright-reasoning-retrieval-first-place-2026-10-28.html</guid><description>FINAL-Bench scores 0.307 nDCG@10 on BRIGHT, Hugging Face&#x27;s new reasoning-intensive retrieval leaderboard. VIDRAFT now leads 12 of 52 official benchmarks.</description><pubDate>2026-10-28T16:30:00+09:00</pubDate></item><item><title>Hugging Face&#x27;s Open-Source Team Built a Free Demo for Darwin-35B-ZTC-Vision</title><link>https://vidraft-ai-media.static.hf.space/a/darwin-35b-ztc-vision-hugging-face-demo-2026-10-28.html</link><guid>https://vidraft-ai-media.static.hf.space/a/darwin-35b-ztc-vision-hugging-face-demo-2026-10-28.html</guid><description>Hugging Face&#x27;s open-source team built a free ZeroGPU demo of Darwin-35B-ZTC-Vision, VIDRAFT&#x27;s 35B vision decision model, and is moving it to FINAL-Bench.</description><pubDate>2026-10-28T15:00:00+09:00</pubDate></item><item><title>Free 1-Minute Website Check for the Age of AI Hacking: Security, SEO and AI Visibility</title><link>https://vidraft-ai-media.static.hf.space/a/freeweb-security-scan-free-website-check-2026-10-27.html</link><guid>https://vidraft-ai-media.static.hf.space/a/freeweb-security-scan-free-website-check-2026-10-27.html</guid><description>FreeWeb Security SCAN gives a read-only report on security headers, HTTPS, email spoofing protection, SEO basics and access for 9 AI crawlers.</description><pubDate>2026-10-27T16:00:00+09:00</pubDate></item><item><title>VKORE-325M: A 325M LLM Pretrained From Zero on a Rebellions ATOM-Max NPU, Equal to a B200 Twin</title><link>https://vidraft-ai-media.static.hf.space/a/vkore-325m-npu-pretraining-atom-max-2026-10-26.html</link><guid>https://vidraft-ai-media.static.hf.space/a/vkore-325m-npu-pretraining-atom-max-2026-10-26.html</guid><description>VIDRAFT pretrained a 325M model from random init on an inference-only Rebellions ATOM-Max NPU. On unseen text: 0.7993 vs 0.7994 for the same model on a B200.</description><pubDate>2026-10-26T16:00:00+09:00</pubDate></item><item><title>Why Long bf16 Training Runs Stall: The Residual Stream Has No Room for Small Updates</title><link>https://vidraft-ai-media.static.hf.space/a/bf16-residual-stream-stall-long-training-2026-10-22.html</link><guid>https://vidraft-ai-media.static.hf.space/a/bf16-residual-stream-stall-long-training-2026-10-22.html</guid><description>A bf16 residual near 600 has a resolution of 4, so small updates vanish. Measured gaps, a reproducible simulation and fixes for long training runs.</description><pubDate>2026-10-22T01:04:34+09:00</pubDate></item><item><title>Darwin-27B-ZTC-v9 Takes #1 on S1MB Across 197 Decision Models</title><link>https://vidraft-ai-media.static.hf.space/a/darwin-27b-ztc-v9-s1mb-number-one-2026-10-20.html</link><guid>https://vidraft-ai-media.static.hf.space/a/darwin-27b-ztc-v9-s1mb-number-one-2026-10-20.html</guid><description>A 27B decision model that answers without generating text now leads the S1MB English leaderboard on Borda score across 137 benchmarks, ahead of entries from Perplexity, Meta and Cloudflare.</description><pubDate>2026-10-20T11:00:00+09:00</pubDate></item><item><title>ZTC: A Decision Model That Answers in One Forward Pass</title><link>https://vidraft-ai-media.static.hf.space/a/ztc-decision-model-one-forward-pass-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/ztc-decision-model-one-forward-pass-2026-10-19.html</guid><description>Darwin-27B-ZTC-v5 returns a calibrated probability for every option without generating text. How decision models work, how they are scored, and how to serve one with transformers serve.</description><pubDate>2026-10-19T15:00:00+09:00</pubDate></item><item><title>LEADBOARD: A Drug Property Benchmark Whose Labels Never Leave the Verifier</title><link>https://vidraft-ai-media.static.hf.space/a/leadboard-held-out-labels-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/leadboard-held-out-labels-2026-10-19.html</guid><description>21 boards across 7 disciplines and 18,382 held-out compounds. Time splits, published noise floors and baselines, and a ladder rule so the board cannot be won by resubmitting.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Open Discovery Challenge: Four Diseases, Four Targets, One Verifier</title><link>https://vidraft-ai-media.static.hf.space/a/open-discovery-challenge-four-seasons-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-discovery-challenge-four-seasons-2026-10-19.html</guid><description>An open drug discovery competition. Submit a SMILES string; a hidden pipeline scores binding, selectivity, ADMET, novelty and synthesis against a target and a counter target.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Open Materials Challenge: Find a Solid-State Battery Electrolyte, Get It Verified</title><link>https://vidraft-ai-media.static.hf.space/a/open-materials-challenge-solid-state-electrolyte-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-materials-challenge-solid-state-electrolyte-2026-10-19.html</guid><description>Submit a lithium-containing crystal structure. A hidden pipeline checks oxidation stability, lithium-metal stability and novelty of use. Season 1 closes 2026-11-30.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Open Quantum Challenge: Two Quantum Seasons, No Quantum Hardware Required</title><link>https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge-gpu-only-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge-gpu-only-2026-10-19.html</guid><description>Compute Fermi-Hubbard ground-state energies or build a surface-code decoder with a GPU and an open harness. Answers are held privately and every entry is scored automatically.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Five Open Science Challenges Where a Verifier, Not the Submitter, Decides the Score</title><link>https://vidraft-ai-media.static.hf.space/a/open-science-challenges-verifier-decides-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-science-challenges-verifier-decides-2026-10-19.html</guid><description>Drug discovery, battery materials, superconductors, quantum computing and drug property prediction: five open challenges from FINAL-Bench, each graded by a verifier that holds the answer privately.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Open Superconductor Challenge: Screen 2D Materials for d-Wave Pairing on a Laptop</title><link>https://vidraft-ai-media.static.hf.space/a/open-superconductor-challenge-laptop-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-superconductor-challenge-laptop-2026-10-19.html</guid><description>Estimate d-wave pairing tendency from a published effective Hubbard model per material. A many-body reference solver verifies the top entries. Season 1 closes 2026-12-31.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Six Open-Source Projects from VIDRAFT, Now on GitHub</title><link>https://vidraft-ai-media.static.hf.space/a/vidraft-open-source-github-six-projects.html</link><guid>https://vidraft-ai-media.static.hf.space/a/vidraft-open-source-github-six-projects.html</guid><description>ONGRID, FINAL Bench, the Open Quantum Challenge harness, POCKET, ZTC and ALL Bench Leaderboard: what each project does and where to start.</description><pubDate>2026-10-18T13:00:00+09:00</pubDate></item><item><title>VIDRAFT This Week: Typed Decisions #1, S1MB Japanese #1, and NPU Pretraining in the Press</title><link>https://vidraft-ai-media.static.hf.space/a/vidraft-week-2026-10-17.html</link><guid>https://vidraft-ai-media.static.hf.space/a/vidraft-week-2026-10-17.html</guid><description>A weekly roundup: Darwin-27B-ZTC-v5 holds #1 on Hugging Face Typed Decisions at 0.81, the S1MB Japanese edition #1 at 84.57, and national press coverage of VKORE pretraining on inference-only NPUs.</description><pubDate>2026-10-17T19:00:00+09:00</pubDate></item><item><title>VIDRAFT Now Leads 11 Hugging Face Official Benchmarks After Reclaiming #1 on Typed Decisions</title><link>https://vidraft-ai-media.static.hf.space/a/typed-decisions-first-eleven-benchmarks.html</link><guid>https://vidraft-ai-media.static.hf.space/a/typed-decisions-first-eleven-benchmarks.html</guid><description>Darwin-27B-ZTC-v5 reclaims #1 on the Typed Decisions official benchmark at 0.81, holding both first and second place, bringing VIDRAFT to #1 on 11 Hugging Face official benchmarks.</description><pubDate>2026-10-16T09:00:00+09:00</pubDate></item><item><title>Pretraining From Scratch on an Inference-Only NPU: VKORE-46M-NPU</title><link>https://vidraft-ai-media.static.hf.space/a/vkore-46m-npu-pretraining-inference-only.html</link><guid>https://vidraft-ai-media.static.hf.space/a/vkore-46m-npu-pretraining-inference-only.html</guid><description>VIDRAFT pretrained a language model from scratch on an inference-only NPU via real backpropagation and released it on Hugging Face. Loss trajectory correlates 0.99 with a GPU reference.</description><pubDate>2026-10-14T16:40:00+09:00</pubDate></item><item><title>Model-Level RSI: What It Means When a Model Improves Its Own Weights</title><link>https://vidraft-ai-media.static.hf.space/a/model-level-rsi-what-it-means.html</link><guid>https://vidraft-ai-media.static.hf.space/a/model-level-rsi-what-it-means.html</guid><description>A practical look at model-level recursive self-improvement, the five stages, and how to read verifiable results like 100% on AIME 2026 and 93.43% on GPQA Diamond.</description><pubDate>2026-10-12T20:10:00+09:00</pubDate></item><item><title>Open Quantum Challenge: Solve Quantum Problems With a GPU and No Quantum Computer</title><link>https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge.html</guid><description>The Open Quantum Challenge on Hugging Face opens with two seasons, quantum simulation and QEC decoding. Use only a GPU and an open harness, no quantum hardware. 2,000 USD in prizes.</description><pubDate>2026-10-11T22:45:00+09:00</pubDate></item><item><title>AI Native: Ten People, Hundreds of AI, and the Arrival of the Pre-AGI Age</title><link>https://vidraft-ai-media.static.hf.space/a/ai-native-vidraft.html</link><guid>https://vidraft-ai-media.static.hf.space/a/ai-native-vidraft.html</guid><description>VIDRAFT is an AI-native lab where hundreds of AI instances run as a company on a few dozen GPUs. Perfect math scores, 165M downloads, quantum hardware, and a self-improving org point to the Pre-AGI age.</description><pubDate>2026-10-11T22:30:00+09:00</pubDate></item><item><title>S1MB Number One: How a Zero-Token Judge Won the Decision-Engine Leaderboard</title><link>https://vidraft-ai-media.static.hf.space/a/s1mb-number-one-decision-engine.html</link><guid>https://vidraft-ai-media.static.hf.space/a/s1mb-number-one-decision-engine.html</guid><description>Our model ranked number one on the S1MB decision-engine leaderboard among 102 models, with a Borda score of 89.58, deciding in a single forward pass with zero generated tokens.</description><pubDate>2026-10-11T22:00:00+09:00</pubDate></item><item><title>a16z Bet $870M on Typed, Machine-Native AI. Here Is the Open, Benchmark-Proven Version</title><link>https://vidraft-ai-media.static.hf.space/a/typed-decision-ai-a16z-vs-open-ztc.html</link><guid>https://vidraft-ai-media.static.hf.space/a/typed-decision-ai-a16z-vs-open-ztc.html</guid><description>TypeSafe AI raised 870M dollars for typed, machine-native AI. We explain what typed decision AI is, why it matters, and show an open, zero-token, S1MB number one alternative.</description><pubDate>2026-10-10T22:00:00+09:00</pubDate></item><item><title>From exams to real work: an open 180B model now leads nine official Hugging Face leaderboards</title><link>https://vidraft-ai-media.static.hf.space/a/open-180b-model-leads-nine-official-hugging-face-leaderboards.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-180b-model-leads-nine-official-hugging-face-leaderboards.html</guid><description>Darwin-180B-RSI adds #1 on IFStruct (structured output, 98.95%) and ExtractBench (document extraction, 90.29), for nine first places among 48 official Hugg</description><pubDate>2026-10-05T21:00:00+09:00</pubDate></item></channel></rss>
 
1
+ <?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>VIDRAFT AI Media</title><link>https://vidraft-ai-media.static.hf.space/</link><description>Engineering news and analysis from VIDRAFT: open LLMs, recursive self-improvement, leaderboards and decision AI, with reproducible numbers.</description><item><title>VIDRAFT in the Korean Press, 30 October 2026: ViSQL, NPU Pretraining, Chitos, a Hugging Face Demo and a Benchmark Analysis</title><link>https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html</link><guid>https://vidraft-ai-media.static.hf.space/a/vidraft-korean-press-roundup-2026-10-30.html</guid><description>Five Korean outlets covered VIDRAFT on 30 October: Chosun Biz on ViSQL, AI Times on the Hugging Face demo, Economist on QuantID&#x27;s benchmark analysis, Financial News on Chitos, and JoongAng Ilbo on NPU pretraining.</description><pubDate>2026-10-30T12:00:00+09:00</pubDate></item><item><title>VIDRAFT NL2SQL Engine Scores 83.80 on BIRD Mini-Dev, Above the Current #1, and Opens a Public Demo</title><link>https://vidraft-ai-media.static.hf.space/a/nl2sql-engine-bird-minidev-83-80-2026-10-29.html</link><guid>https://vidraft-ai-media.static.hf.space/a/nl2sql-engine-bird-minidev-83-80-2026-10-29.html</guid><description>Darwin-180B-NL2SQL reaches 83.80% execution accuracy on BIRD Mini-Dev (SQLite), above the listed #1 at 81.00. A public demo drafts a legal fraud notice from data.</description><pubDate>2026-10-29T18:00:00+09:00</pubDate></item><item><title>Chitos Covers All 17 Items of Korea&#x27;s Financial AI Agent Security Criteria, Scores 77.8 on OWASP Benchmark On-Prem</title><link>https://vidraft-ai-media.static.hf.space/a/chitos-fsi-17-owasp-77-8-2026-10-29.html</link><guid>https://vidraft-ai-media.static.hf.space/a/chitos-fsi-17-owasp-77-8-2026-10-29.html</guid><description>VIDRAFT&#x27;s Chitos has modules for all 17 items of Korea&#x27;s financial AI agent security criteria and scores 77.8 (Youden) on OWASP Benchmark with an on-premises engine.</description><pubDate>2026-10-29T17:40:00+09:00</pubDate></item><item><title>Decision AI, Explained: Why Hugging Face Built a Demo for Darwin-35B-ZTC-Vision</title><link>https://vidraft-ai-media.static.hf.space/a/decision-ai-ztc-vision-records-2026-10-29.html</link><guid>https://vidraft-ai-media.static.hf.space/a/decision-ai-ztc-vision-records-2026-10-29.html</guid><description>Darwin-35B-ZTC-Vision answers with a calibrated probability per option in one pass. Hugging Face&#x27;s open-source team built and donated its demo. Records inside.</description><pubDate>2026-10-29T17:20:00+09:00</pubDate></item><item><title>VIDRAFT Takes #1 on BRIGHT, the New Reasoning-Retrieval Leaderboard on Hugging Face</title><link>https://vidraft-ai-media.static.hf.space/a/bright-reasoning-retrieval-first-place-2026-10-28.html</link><guid>https://vidraft-ai-media.static.hf.space/a/bright-reasoning-retrieval-first-place-2026-10-28.html</guid><description>FINAL-Bench scores 0.307 nDCG@10 on BRIGHT, Hugging Face&#x27;s new reasoning-intensive retrieval leaderboard. VIDRAFT now leads 12 of 52 official benchmarks.</description><pubDate>2026-10-28T16:30:00+09:00</pubDate></item><item><title>Hugging Face&#x27;s Open-Source Team Built a Free Demo for Darwin-35B-ZTC-Vision</title><link>https://vidraft-ai-media.static.hf.space/a/darwin-35b-ztc-vision-hugging-face-demo-2026-10-28.html</link><guid>https://vidraft-ai-media.static.hf.space/a/darwin-35b-ztc-vision-hugging-face-demo-2026-10-28.html</guid><description>Hugging Face&#x27;s open-source team built a free ZeroGPU demo of Darwin-35B-ZTC-Vision, VIDRAFT&#x27;s 35B vision decision model, and is moving it to FINAL-Bench.</description><pubDate>2026-10-28T15:00:00+09:00</pubDate></item><item><title>Free 1-Minute Website Check for the Age of AI Hacking: Security, SEO and AI Visibility</title><link>https://vidraft-ai-media.static.hf.space/a/freeweb-security-scan-free-website-check-2026-10-27.html</link><guid>https://vidraft-ai-media.static.hf.space/a/freeweb-security-scan-free-website-check-2026-10-27.html</guid><description>FreeWeb Security SCAN gives a read-only report on security headers, HTTPS, email spoofing protection, SEO basics and access for 9 AI crawlers.</description><pubDate>2026-10-27T16:00:00+09:00</pubDate></item><item><title>VKORE-325M: A 325M LLM Pretrained From Zero on a Rebellions ATOM-Max NPU, Equal to a B200 Twin</title><link>https://vidraft-ai-media.static.hf.space/a/vkore-325m-npu-pretraining-atom-max-2026-10-26.html</link><guid>https://vidraft-ai-media.static.hf.space/a/vkore-325m-npu-pretraining-atom-max-2026-10-26.html</guid><description>VIDRAFT pretrained a 325M model from random init on an inference-only Rebellions ATOM-Max NPU. On unseen text: 0.7993 vs 0.7994 for the same model on a B200.</description><pubDate>2026-10-26T16:00:00+09:00</pubDate></item><item><title>Why Long bf16 Training Runs Stall: The Residual Stream Has No Room for Small Updates</title><link>https://vidraft-ai-media.static.hf.space/a/bf16-residual-stream-stall-long-training-2026-10-22.html</link><guid>https://vidraft-ai-media.static.hf.space/a/bf16-residual-stream-stall-long-training-2026-10-22.html</guid><description>A bf16 residual near 600 has a resolution of 4, so small updates vanish. Measured gaps, a reproducible simulation and fixes for long training runs.</description><pubDate>2026-10-22T01:04:34+09:00</pubDate></item><item><title>Darwin-27B-ZTC-v9 Takes #1 on S1MB Across 197 Decision Models</title><link>https://vidraft-ai-media.static.hf.space/a/darwin-27b-ztc-v9-s1mb-number-one-2026-10-20.html</link><guid>https://vidraft-ai-media.static.hf.space/a/darwin-27b-ztc-v9-s1mb-number-one-2026-10-20.html</guid><description>A 27B decision model that answers without generating text now leads the S1MB English leaderboard on Borda score across 137 benchmarks, ahead of entries from Perplexity, Meta and Cloudflare.</description><pubDate>2026-10-20T11:00:00+09:00</pubDate></item><item><title>ZTC: A Decision Model That Answers in One Forward Pass</title><link>https://vidraft-ai-media.static.hf.space/a/ztc-decision-model-one-forward-pass-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/ztc-decision-model-one-forward-pass-2026-10-19.html</guid><description>Darwin-27B-ZTC-v5 returns a calibrated probability for every option without generating text. How decision models work, how they are scored, and how to serve one with transformers serve.</description><pubDate>2026-10-19T15:00:00+09:00</pubDate></item><item><title>LEADBOARD: A Drug Property Benchmark Whose Labels Never Leave the Verifier</title><link>https://vidraft-ai-media.static.hf.space/a/leadboard-held-out-labels-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/leadboard-held-out-labels-2026-10-19.html</guid><description>21 boards across 7 disciplines and 18,382 held-out compounds. Time splits, published noise floors and baselines, and a ladder rule so the board cannot be won by resubmitting.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Open Discovery Challenge: Four Diseases, Four Targets, One Verifier</title><link>https://vidraft-ai-media.static.hf.space/a/open-discovery-challenge-four-seasons-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-discovery-challenge-four-seasons-2026-10-19.html</guid><description>An open drug discovery competition. Submit a SMILES string; a hidden pipeline scores binding, selectivity, ADMET, novelty and synthesis against a target and a counter target.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Open Materials Challenge: Find a Solid-State Battery Electrolyte, Get It Verified</title><link>https://vidraft-ai-media.static.hf.space/a/open-materials-challenge-solid-state-electrolyte-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-materials-challenge-solid-state-electrolyte-2026-10-19.html</guid><description>Submit a lithium-containing crystal structure. A hidden pipeline checks oxidation stability, lithium-metal stability and novelty of use. Season 1 closes 2026-11-30.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Open Quantum Challenge: Two Quantum Seasons, No Quantum Hardware Required</title><link>https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge-gpu-only-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge-gpu-only-2026-10-19.html</guid><description>Compute Fermi-Hubbard ground-state energies or build a surface-code decoder with a GPU and an open harness. Answers are held privately and every entry is scored automatically.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Five Open Science Challenges Where a Verifier, Not the Submitter, Decides the Score</title><link>https://vidraft-ai-media.static.hf.space/a/open-science-challenges-verifier-decides-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-science-challenges-verifier-decides-2026-10-19.html</guid><description>Drug discovery, battery materials, superconductors, quantum computing and drug property prediction: five open challenges from FINAL-Bench, each graded by a verifier that holds the answer privately.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Open Superconductor Challenge: Screen 2D Materials for d-Wave Pairing on a Laptop</title><link>https://vidraft-ai-media.static.hf.space/a/open-superconductor-challenge-laptop-2026-10-19.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-superconductor-challenge-laptop-2026-10-19.html</guid><description>Estimate d-wave pairing tendency from a published effective Hubbard model per material. A many-body reference solver verifies the top entries. Season 1 closes 2026-12-31.</description><pubDate>2026-10-19T10:00:00+09:00</pubDate></item><item><title>Six Open-Source Projects from VIDRAFT, Now on GitHub</title><link>https://vidraft-ai-media.static.hf.space/a/vidraft-open-source-github-six-projects.html</link><guid>https://vidraft-ai-media.static.hf.space/a/vidraft-open-source-github-six-projects.html</guid><description>ONGRID, FINAL Bench, the Open Quantum Challenge harness, POCKET, ZTC and ALL Bench Leaderboard: what each project does and where to start.</description><pubDate>2026-10-18T13:00:00+09:00</pubDate></item><item><title>VIDRAFT This Week: Typed Decisions #1, S1MB Japanese #1, and NPU Pretraining in the Press</title><link>https://vidraft-ai-media.static.hf.space/a/vidraft-week-2026-10-17.html</link><guid>https://vidraft-ai-media.static.hf.space/a/vidraft-week-2026-10-17.html</guid><description>A weekly roundup: Darwin-27B-ZTC-v5 holds #1 on Hugging Face Typed Decisions at 0.81, the S1MB Japanese edition #1 at 84.57, and national press coverage of VKORE pretraining on inference-only NPUs.</description><pubDate>2026-10-17T19:00:00+09:00</pubDate></item><item><title>VIDRAFT Now Leads 11 Hugging Face Official Benchmarks After Reclaiming #1 on Typed Decisions</title><link>https://vidraft-ai-media.static.hf.space/a/typed-decisions-first-eleven-benchmarks.html</link><guid>https://vidraft-ai-media.static.hf.space/a/typed-decisions-first-eleven-benchmarks.html</guid><description>Darwin-27B-ZTC-v5 reclaims #1 on the Typed Decisions official benchmark at 0.81, holding both first and second place, bringing VIDRAFT to #1 on 11 Hugging Face official benchmarks.</description><pubDate>2026-10-16T09:00:00+09:00</pubDate></item><item><title>Pretraining From Scratch on an Inference-Only NPU: VKORE-46M-NPU</title><link>https://vidraft-ai-media.static.hf.space/a/vkore-46m-npu-pretraining-inference-only.html</link><guid>https://vidraft-ai-media.static.hf.space/a/vkore-46m-npu-pretraining-inference-only.html</guid><description>VIDRAFT pretrained a language model from scratch on an inference-only NPU via real backpropagation and released it on Hugging Face. Loss trajectory correlates 0.99 with a GPU reference.</description><pubDate>2026-10-14T16:40:00+09:00</pubDate></item><item><title>Model-Level RSI: What It Means When a Model Improves Its Own Weights</title><link>https://vidraft-ai-media.static.hf.space/a/model-level-rsi-what-it-means.html</link><guid>https://vidraft-ai-media.static.hf.space/a/model-level-rsi-what-it-means.html</guid><description>A practical look at model-level recursive self-improvement, the five stages, and how to read verifiable results like 100% on AIME 2026 and 93.43% on GPQA Diamond.</description><pubDate>2026-10-12T20:10:00+09:00</pubDate></item><item><title>Open Quantum Challenge: Solve Quantum Problems With a GPU and No Quantum Computer</title><link>https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-quantum-challenge.html</guid><description>The Open Quantum Challenge on Hugging Face opens with two seasons, quantum simulation and QEC decoding. Use only a GPU and an open harness, no quantum hardware. 2,000 USD in prizes.</description><pubDate>2026-10-11T22:45:00+09:00</pubDate></item><item><title>AI Native: Ten People, Hundreds of AI, and the Arrival of the Pre-AGI Age</title><link>https://vidraft-ai-media.static.hf.space/a/ai-native-vidraft.html</link><guid>https://vidraft-ai-media.static.hf.space/a/ai-native-vidraft.html</guid><description>VIDRAFT is an AI-native lab where hundreds of AI instances run as a company on a few dozen GPUs. Perfect math scores, 165M downloads, quantum hardware, and a self-improving org point to the Pre-AGI age.</description><pubDate>2026-10-11T22:30:00+09:00</pubDate></item><item><title>S1MB Number One: How a Zero-Token Judge Won the Decision-Engine Leaderboard</title><link>https://vidraft-ai-media.static.hf.space/a/s1mb-number-one-decision-engine.html</link><guid>https://vidraft-ai-media.static.hf.space/a/s1mb-number-one-decision-engine.html</guid><description>Our model ranked number one on the S1MB decision-engine leaderboard among 102 models, with a Borda score of 89.58, deciding in a single forward pass with zero generated tokens.</description><pubDate>2026-10-11T22:00:00+09:00</pubDate></item><item><title>a16z Bet $870M on Typed, Machine-Native AI. Here Is the Open, Benchmark-Proven Version</title><link>https://vidraft-ai-media.static.hf.space/a/typed-decision-ai-a16z-vs-open-ztc.html</link><guid>https://vidraft-ai-media.static.hf.space/a/typed-decision-ai-a16z-vs-open-ztc.html</guid><description>TypeSafe AI raised 870M dollars for typed, machine-native AI. We explain what typed decision AI is, why it matters, and show an open, zero-token, S1MB number one alternative.</description><pubDate>2026-10-10T22:00:00+09:00</pubDate></item><item><title>From exams to real work: an open 180B model now leads nine official Hugging Face leaderboards</title><link>https://vidraft-ai-media.static.hf.space/a/open-180b-model-leads-nine-official-hugging-face-leaderboards.html</link><guid>https://vidraft-ai-media.static.hf.space/a/open-180b-model-leads-nine-official-hugging-face-leaderboards.html</guid><description>Darwin-180B-RSI adds #1 on IFStruct (structured output, 98.95%) and ExtractBench (document extraction, 90.29), for nine first places among 48 official Hugg</description><pubDate>2026-10-05T21:00:00+09:00</pubDate></item></channel></rss>