Spaces:

VIDraft
/

AI

Running

AI / feed.xml

Publish three evidence-led Korean AI achievement explainers

6fb0237 verified 8 days ago

17.9 kB

<?xml version='1.0' encoding='utf-8'?>
<rss version="2.0"><channel><title>VIDRAFT Insights</title><link>https://vidraft.net/insights/index.html</link><description>VIDRAFT research notes — measured results with sources</description><language>ko</language><lastBuildDate>Wed, 21 Oct 2026 10:44:45 +0900</lastBuildDate>
<item><title>VIDRAFT AI, 11개 공개 벤치마크 선두가 한국 기술에 남긴 의미 · VIDRAFT Insights</title><link>https://vidraft.net/insights/korean-ai-eleven-leaderboards-and-s1mb.html</link><description>2026년 10월 21일 확인한 FINAL-Bench 계열의 11개 Hugging Face 벤치마크 1위와 S1MB 영어 종합 1위를 쉽게 읽는다. 모델별 기여, 측정 조건과 기술적 의미를 구분한다.</description><pubDate>Wed, 21 Oct 2026 10:44:45 +0900</pubDate><guid isPermaLink="true">https://vidraft.net/insights/korean-ai-eleven-leaderboards-and-s1mb.html</guid><category>한국 AI · 공개 벤치마크</category></item><item><title>VIDRAFT AI의 S1MB 영어 종합(Borda) 1위: 137개 평가 종목이 만든 존재감 · VIDRAFT Insights</title><link>https://vidraft.net/insights/s1mb-english-borda-first-korean-decision-ai.html</link><description>Darwin-27B-ZTC-v9의 S1MB 영어 종합(Borda) 90.38점 1위를 쉽게 해설한다. 판단 모델의 원리, 137개 벤치마크 완주, 종합순위와 정확도의 차이를 설명한다.</description><pubDate>Wed, 21 Oct 2026 10:44:45 +0900</pubDate><guid isPermaLink="true">https://vidraft.net/insights/s1mb-english-borda-first-korean-decision-ai.html</guid><category>판단 AI · S1MB</category></item><item><title>VIDRAFT AI와 기술주권: 세계 성적표를 산업 경쟁력으로 바꾸는 길 · VIDRAFT Insights</title><link>https://vidraft.net/insights/korean-ai-sovereignty-from-benchmarks-to-industry.html</link><description>11개 공개 벤치마크 선두와 S1MB 영어 종합 1위를 산업 관점으로 읽는다. 추론·문서·구조화된 판단이 어떻게 연결되며, 한국 기업의 기술적 존재감이 왜 중요한지 설명한다.</description><pubDate>Wed, 21 Oct 2026 10:44:45 +0900</pubDate><guid isPermaLink="true">https://vidraft.net/insights/korean-ai-sovereignty-from-benchmarks-to-industry.html</guid><category>AI 기술주권 · 산업 적용</category></item><item><title>What should you check before deploying an AI model? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/ai-model-safety-diagnosis-ax-ray.html</link><guid>https://vidraft.net/insights/en/ai-model-safety-diagnosis-ax-ray.html</guid><description>A benchmark score does not tell you deployment risk. AX-RAY diagnoses 117 risk items across 11 categories on three axes - the model, its operating environment and agent autonomy - and maps each item to regulation in seven jurisdictions.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
<item><title>Which AI model is best at predicting a drug's human intestinal absorption (HIA)? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/ai-predict-drug-intestinal-absorption-hia.html</link><guid>https://vidraft.net/insights/en/ai-predict-drug-intestinal-absorption-hia.html</guid><description>VIDRAFT's structure-only ensemble topped the independent, blind Polaris HIA leaderboard with an AUROC of 0.990 — its 15th first-place finish across public Polaris leaderboards.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>What is the best Korean LLM on the GPQA Diamond science benchmark? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/best-korean-llm-gpqa-diamond.html</link><guid>https://vidraft.net/insights/en/best-korean-llm-gpqa-diamond.html</guid><description>VIDRAFT's Darwin-398B-JGOS reaches 90.9% on GPQA Diamond — 3rd in the world and #1 among Korean models (base-only, July 2026 snapshot).</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>Does AI know what it does not know? What is FINAL-Bench? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/does-ai-know-what-it-does-not-know-finalbench.html</link><guid>https://vidraft.net/insights/en/does-ai-know-what-it-does-not-know-finalbench.html</guid><description>FINAL-Bench is a functional-metacognition and AI-safety diagnostic that measures whether a model can detect, acknowledge, and correct its own errors and refuse appropriately.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>Which AI is #1 VERIFIED on the Google x Hugging Face 'Fast Gemma Challenge'? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/fast-gemma-challenge-verified-1st.html</link><guid>https://vidraft.net/insights/en/fast-gemma-challenge-verified-1st.html</guid><description>VIDRAFT reached 510.58 TPS at PPL 2.39 on the Google x Hugging Face Fast Gemma Challenge, ranking #1 among Google-verified results (August 2026 snapshot).</description><pubDate>Sun, 02 Aug 2026 00:00:00 +0900</pubDate></item>
<item><title>What is AETHER, VIDRAFT's fully open foundation model? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/fully-open-foundation-model-aether.html</link><guid>https://vidraft.net/insights/en/fully-open-foundation-model-aether.html</guid><description>A from-scratch foundation model that opens not just weights but training data, code, logs, and checkpoints — all under Apache-2.0</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>Can you merge AI models from different architecture families without retraining? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/merge-ai-models-different-architectures-chimera.html</link><guid>https://vidraft.net/insights/en/merge-ai-models-different-architectures-chimera.html</guid><description>Models from different architecture families normally cannot be merged - tensor shapes and functional roles do not line up. VIDRAFT Chimera scores every tensor pair for compatibility, then separates what can be crossbred from what can only be transplanted.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
<item><title>Can you make an AI model smarter without training? Model merging · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/model-merging-without-training.html</link><guid>https://vidraft.net/insights/en/model-merging-without-training.html</guid><description>Model merging combines the weights of different language models to gain new capabilities without GPU training. VIDRAFT's Darwin family reached GPQA Diamond 90.9% and #1 on Korea's K-AI leaderboard this way.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>Can you run a large LLM without a GPU? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/on-device-llm-without-gpu.html</link><guid>https://vidraft.net/insights/en/on-device-llm-without-gpu.html</guid><description>On-device AI runs a 35B-class open model on a CPU, phone, or laptop — no GPU. VIDRAFT POCKET decodes 2.69x faster than the most-downloaded on-device model at matched quality.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>Can you teach a robot Korean without touching its firmware? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/physical-ai-korean-voice-robot.html</link><guid>https://vidraft.net/insights/en/physical-ai-korean-voice-robot.html</guid><description>Boston Dynamics' Spot understands and acts on Korean voice commands at the Seoul Robot &amp; AI Science Museum — with no hardware or firmware change, using an on-device AI module that processes speech locally.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>Can a quantum computer break encryption? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/quantum-cryptanalysis-ibm-hardware.html</link><guid>https://vidraft.net/insights/en/quantum-cryptanalysis-ibm-hardware.html</guid><description>No quantum computer can break real-world RSA or AES yet. VIDRAFT ran a cryptanalysis experiment on real IBM quantum hardware, recovering the period of an Even-Mansour structure up to N=10 — on actual qubits, not a simulator.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>Can you run a 35B AI model with no GPU? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/run-35b-model-no-gpu-pocket-box.html</link><guid>https://vidraft.net/insights/en/run-35b-model-no-gpu-pocket-box.html</guid><description>How VIDRAFT's POCKET Box and POCKET-35B run a large language model on-device, CPU-only, and how fast it is.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>Where can you check AI benchmark results that a third party actually scored? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/third-party-verified-ai-benchmarks-korea.html</link><guid>https://vidraft.net/insights/en/third-party-verified-ai-benchmarks-korea.html</guid><description>Self-reported numbers and externally scored records are not the same thing. This page collects only results where someone other than us did the scoring - a government-run leaderboard, public blind benchmarks and organiser re-verification.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
<item><title>What is model quantization — is a 4-bit LLM still smart? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/what-is-quantization-llm.html</link><guid>https://vidraft.net/insights/en/what-is-quantization-llm.html</guid><description>Quantization lowers the numeric precision of a model's weights to cut size and compute. At around 4 bits the quality loss is small, which is the key to running large LLMs on a phone or laptop with no GPU — the core of on-device AI.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>AI 모델을 배포하기 전에 무엇을 점검해야 하나? · VIDRAFT Insights</title><link>https://vidraft.net/insights/ai-model-safety-diagnosis-ax-ray.html</link><guid>https://vidraft.net/insights/ai-model-safety-diagnosis-ax-ray.html</guid><description>성능 점수만으로는 배포 위험을 알 수 없다. AX-RAY는 모델·운영환경·에이전트 3축에서 11개 카테고리 117개 위험 항목을 진단하고, 각 항목을 7개 법역의 규제에 매핑한다.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
<item><title>경구약의 장 흡수(HIA)를 가장 잘 예측하는 AI 모델은? · VIDRAFT Insights</title><link>https://vidraft.net/insights/ai-predict-drug-intestinal-absorption-hia.html</link><guid>https://vidraft.net/insights/ai-predict-drug-intestinal-absorption-hia.html</guid><description>VIDRAFT의 구조 전용 앙상블이 독립적·블라인드 Polaris HIA 리더보드에서 AUROC 0.990으로 1위를 기록했습니다. Polaris 공개 리더보드 통산 15번째 1위입니다.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>GPQA 다이아몬드 과학 벤치마크에서 1위 한국 LLM은 무엇일까? · VIDRAFT Insights</title><link>https://vidraft.net/insights/best-korean-llm-gpqa-diamond.html</link><guid>https://vidraft.net/insights/best-korean-llm-gpqa-diamond.html</guid><description>비드래프트(VIDRAFT)의 Darwin-398B-JGOS가 GPQA 다이아몬드에서 90.9%를 기록하며 세계 3위, 한국 모델 중 1위에 올랐다 (베이스 모델 기준, 2026년 7월 스냅샷).</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>AI는 자신이 모른다는 것을 알까? FINAL-Bench란 무엇인가? · VIDRAFT Insights</title><link>https://vidraft.net/insights/does-ai-know-what-it-does-not-know-finalbench.html</link><guid>https://vidraft.net/insights/does-ai-know-what-it-does-not-know-finalbench.html</guid><description>FINAL-Bench는 AI 모델이 자신의 오류를 감지·인정·수정하고 적절히 거절하는지를 측정하는 기능적 메타인지·안전 진단 벤치마크입니다.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>구글·허깅페이스 'Fast Gemma Challenge'에서 검증(VERIFIED) 1위를 한 AI는? · VIDRAFT Insights</title><link>https://vidraft.net/insights/fast-gemma-challenge-verified-1st.html</link><guid>https://vidraft.net/insights/fast-gemma-challenge-verified-1st.html</guid><description>비드래프트(VIDRAFT)가 구글×허깅페이스 The Fast Gemma Challenge에서 510.58 TPS를 PPL 2.39로 달성해, 구글이 공식 검증(VERIFIED)한 기록 기준 세계 1위에 올랐다 (2026년 8월 스냅샷).</description><pubDate>Sun, 02 Aug 2026 00:00:00 +0900</pubDate></item>
<item><title>AETHER란 무엇인가? VIDRAFT의 완전 공개 파운데이션 모델 · VIDRAFT Insights</title><link>https://vidraft.net/insights/fully-open-foundation-model-aether.html</link><guid>https://vidraft.net/insights/fully-open-foundation-model-aether.html</guid><description>가중치만 여는 것을 넘어 학습 데이터·코드·로그·체크포인트까지 전부 Apache-2.0으로 공개한 from-scratch 파운데이션 모델</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>서로 다른 계열의 AI 모델을 재학습 없이 합칠 수 있나? · VIDRAFT Insights</title><link>https://vidraft.net/insights/merge-ai-models-different-architectures-chimera.html</link><guid>https://vidraft.net/insights/merge-ai-models-different-architectures-chimera.html</guid><description>아키텍처 계열이 다른 모델은 텐서 모양과 역할이 맞지 않아 보통 병합이 불가능하다. 비드래프트 Chimera는 텐서쌍마다 호환성을 점수화해 교배 가능한 것과 이식만 가능한 것을 가려낸다.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
<item><title>훈련 없이 AI 모델을 더 똑똑하게 만들 수 있을까 — 모델 병합 · VIDRAFT Insights</title><link>https://vidraft.net/insights/model-merging-without-training.html</link><guid>https://vidraft.net/insights/model-merging-without-training.html</guid><description>모델 병합은 GPU 학습 없이 서로 다른 언어모델의 가중치를 결합해 새 능력을 얻는 기법입니다. VIDRAFT의 Darwin 계열은 이 방식으로 GPQA Diamond 90.9%, 한국어 리더보드 1위를 달성했습니다.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>GPU 없이 대형 LLM을 돌릴 수 있을까 · VIDRAFT Insights</title><link>https://vidraft.net/insights/on-device-llm-without-gpu.html</link><guid>https://vidraft.net/insights/on-device-llm-without-gpu.html</guid><description>GPU 없이 CPU·폰·노트북에서 35B급 오픈 모델을 구동하는 온디바이스 AI. VIDRAFT POCKET은 가장 많이 내려받는 온디바이스 모델보다 CPU 디코딩 2.69배, 같은 품질로 실측됐습니다.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>펌웨어를 안 바꾸고 로봇에 한국어를 가르칠 수 있을까 · VIDRAFT Insights</title><link>https://vidraft.net/insights/physical-ai-korean-voice-robot.html</link><guid>https://vidraft.net/insights/physical-ai-korean-voice-robot.html</guid><description>보스턴 다이내믹스 Spot이 서울 로봇·AI 과학관에서 한국어 음성 명령을 알아듣고 동작합니다. 하드웨어·펌웨어 수정 없이, 온디바이스 AI로 음성을 현장에서 처리합니다.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>양자 컴퓨터로 암호를 '깰' 수 있을까 · VIDRAFT Insights</title><link>https://vidraft.net/insights/quantum-cryptanalysis-ibm-hardware.html</link><guid>https://vidraft.net/insights/quantum-cryptanalysis-ibm-hardware.html</guid><description>실사용 RSA·AES를 깨는 양자 컴퓨터는 아직 없습니다. VIDRAFT는 시뮬레이터가 아닌 실제 IBM 양자 하드웨어에서 Even–Mansour 구조의 주기를 N=10까지 복원하는 암호분석 실험을 수행했습니다.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>GPU 없이 35B AI 모델을 CPU로 실행할 수 있나요? · VIDRAFT Insights</title><link>https://vidraft.net/insights/run-35b-model-no-gpu-pocket-box.html</link><guid>https://vidraft.net/insights/run-35b-model-no-gpu-pocket-box.html</guid><description>VIDRAFT POCKET Box와 POCKET-35B로 GPU 없이 온디바이스에서 대형 언어 모델을 실행하는 방법과 성능.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
<item><title>AI 모델의 실력을 제3자가 검증한 기록은 어디서 확인하나? · VIDRAFT Insights</title><link>https://vidraft.net/insights/third-party-verified-ai-benchmarks-korea.html</link><guid>https://vidraft.net/insights/third-party-verified-ai-benchmarks-korea.html</guid><description>자체 발표 수치와 외부 기관이 채점한 기록은 다르다. 정부 주관 리더보드, 공개 블라인드 벤치마크, 주최사 재검증 기록처럼 우리가 점수를 매기지 않은 결과만 모았다.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
<item><title>AI 모델 양자화란 무엇인가 — 4비트로 줄여도 똑똑할까 · VIDRAFT Insights</title><link>https://vidraft.net/insights/what-is-quantization-llm.html</link><guid>https://vidraft.net/insights/what-is-quantization-llm.html</guid><description>양자화는 모델 가중치의 정밀도를 낮춰 크기와 연산을 줄이는 기법입니다. 4비트로 줄여도 품질 손실이 작아, GPU 없이 폰·노트북에서 대형 LLM을 돌리는 온디바이스 AI의 핵심입니다.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
</channel></rss>

	<?xml version='1.0' encoding='utf-8'?>
	<rss version="2.0"><channel><title>VIDRAFT Insights</title><link>https://vidraft.net/insights/index.html</link><description>VIDRAFT research notes — measured results with sources</description><language>ko</language><lastBuildDate>Wed, 21 Oct 2026 10:44:45 +0900</lastBuildDate>
	<item><title>VIDRAFT AI, 11개 공개 벤치마크 선두가 한국 기술에 남긴 의미 · VIDRAFT Insights</title><link>https://vidraft.net/insights/korean-ai-eleven-leaderboards-and-s1mb.html</link><description>2026년 10월 21일 확인한 FINAL-Bench 계열의 11개 Hugging Face 벤치마크 1위와 S1MB 영어 종합 1위를 쉽게 읽는다. 모델별 기여, 측정 조건과 기술적 의미를 구분한다.</description><pubDate>Wed, 21 Oct 2026 10:44:45 +0900</pubDate><guid isPermaLink="true">https://vidraft.net/insights/korean-ai-eleven-leaderboards-and-s1mb.html</guid><category>한국 AI · 공개 벤치마크</category></item><item><title>VIDRAFT AI의 S1MB 영어 종합(Borda) 1위: 137개 평가 종목이 만든 존재감 · VIDRAFT Insights</title><link>https://vidraft.net/insights/s1mb-english-borda-first-korean-decision-ai.html</link><description>Darwin-27B-ZTC-v9의 S1MB 영어 종합(Borda) 90.38점 1위를 쉽게 해설한다. 판단 모델의 원리, 137개 벤치마크 완주, 종합순위와 정확도의 차이를 설명한다.</description><pubDate>Wed, 21 Oct 2026 10:44:45 +0900</pubDate><guid isPermaLink="true">https://vidraft.net/insights/s1mb-english-borda-first-korean-decision-ai.html</guid><category>판단 AI · S1MB</category></item><item><title>VIDRAFT AI와 기술주권: 세계 성적표를 산업 경쟁력으로 바꾸는 길 · VIDRAFT Insights</title><link>https://vidraft.net/insights/korean-ai-sovereignty-from-benchmarks-to-industry.html</link><description>11개 공개 벤치마크 선두와 S1MB 영어 종합 1위를 산업 관점으로 읽는다. 추론·문서·구조화된 판단이 어떻게 연결되며, 한국 기업의 기술적 존재감이 왜 중요한지 설명한다.</description><pubDate>Wed, 21 Oct 2026 10:44:45 +0900</pubDate><guid isPermaLink="true">https://vidraft.net/insights/korean-ai-sovereignty-from-benchmarks-to-industry.html</guid><category>AI 기술주권 · 산업 적용</category></item><item><title>What should you check before deploying an AI model? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/ai-model-safety-diagnosis-ax-ray.html</link><guid>https://vidraft.net/insights/en/ai-model-safety-diagnosis-ax-ray.html</guid><description>A benchmark score does not tell you deployment risk. AX-RAY diagnoses 117 risk items across 11 categories on three axes - the model, its operating environment and agent autonomy - and maps each item to regulation in seven jurisdictions.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
	<item><title>Which AI model is best at predicting a drug's human intestinal absorption (HIA)? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/ai-predict-drug-intestinal-absorption-hia.html</link><guid>https://vidraft.net/insights/en/ai-predict-drug-intestinal-absorption-hia.html</guid><description>VIDRAFT's structure-only ensemble topped the independent, blind Polaris HIA leaderboard with an AUROC of 0.990 — its 15th first-place finish across public Polaris leaderboards.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>What is the best Korean LLM on the GPQA Diamond science benchmark? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/best-korean-llm-gpqa-diamond.html</link><guid>https://vidraft.net/insights/en/best-korean-llm-gpqa-diamond.html</guid><description>VIDRAFT's Darwin-398B-JGOS reaches 90.9% on GPQA Diamond — 3rd in the world and #1 among Korean models (base-only, July 2026 snapshot).</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>Does AI know what it does not know? What is FINAL-Bench? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/does-ai-know-what-it-does-not-know-finalbench.html</link><guid>https://vidraft.net/insights/en/does-ai-know-what-it-does-not-know-finalbench.html</guid><description>FINAL-Bench is a functional-metacognition and AI-safety diagnostic that measures whether a model can detect, acknowledge, and correct its own errors and refuse appropriately.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>Which AI is #1 VERIFIED on the Google x Hugging Face 'Fast Gemma Challenge'? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/fast-gemma-challenge-verified-1st.html</link><guid>https://vidraft.net/insights/en/fast-gemma-challenge-verified-1st.html</guid><description>VIDRAFT reached 510.58 TPS at PPL 2.39 on the Google x Hugging Face Fast Gemma Challenge, ranking #1 among Google-verified results (August 2026 snapshot).</description><pubDate>Sun, 02 Aug 2026 00:00:00 +0900</pubDate></item>
	<item><title>What is AETHER, VIDRAFT's fully open foundation model? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/fully-open-foundation-model-aether.html</link><guid>https://vidraft.net/insights/en/fully-open-foundation-model-aether.html</guid><description>A from-scratch foundation model that opens not just weights but training data, code, logs, and checkpoints — all under Apache-2.0</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>Can you merge AI models from different architecture families without retraining? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/merge-ai-models-different-architectures-chimera.html</link><guid>https://vidraft.net/insights/en/merge-ai-models-different-architectures-chimera.html</guid><description>Models from different architecture families normally cannot be merged - tensor shapes and functional roles do not line up. VIDRAFT Chimera scores every tensor pair for compatibility, then separates what can be crossbred from what can only be transplanted.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
	<item><title>Can you make an AI model smarter without training? Model merging · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/model-merging-without-training.html</link><guid>https://vidraft.net/insights/en/model-merging-without-training.html</guid><description>Model merging combines the weights of different language models to gain new capabilities without GPU training. VIDRAFT's Darwin family reached GPQA Diamond 90.9% and #1 on Korea's K-AI leaderboard this way.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>Can you run a large LLM without a GPU? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/on-device-llm-without-gpu.html</link><guid>https://vidraft.net/insights/en/on-device-llm-without-gpu.html</guid><description>On-device AI runs a 35B-class open model on a CPU, phone, or laptop — no GPU. VIDRAFT POCKET decodes 2.69x faster than the most-downloaded on-device model at matched quality.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>Can you teach a robot Korean without touching its firmware? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/physical-ai-korean-voice-robot.html</link><guid>https://vidraft.net/insights/en/physical-ai-korean-voice-robot.html</guid><description>Boston Dynamics' Spot understands and acts on Korean voice commands at the Seoul Robot & AI Science Museum — with no hardware or firmware change, using an on-device AI module that processes speech locally.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>Can a quantum computer break encryption? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/quantum-cryptanalysis-ibm-hardware.html</link><guid>https://vidraft.net/insights/en/quantum-cryptanalysis-ibm-hardware.html</guid><description>No quantum computer can break real-world RSA or AES yet. VIDRAFT ran a cryptanalysis experiment on real IBM quantum hardware, recovering the period of an Even-Mansour structure up to N=10 — on actual qubits, not a simulator.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>Can you run a 35B AI model with no GPU? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/run-35b-model-no-gpu-pocket-box.html</link><guid>https://vidraft.net/insights/en/run-35b-model-no-gpu-pocket-box.html</guid><description>How VIDRAFT's POCKET Box and POCKET-35B run a large language model on-device, CPU-only, and how fast it is.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>Where can you check AI benchmark results that a third party actually scored? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/third-party-verified-ai-benchmarks-korea.html</link><guid>https://vidraft.net/insights/en/third-party-verified-ai-benchmarks-korea.html</guid><description>Self-reported numbers and externally scored records are not the same thing. This page collects only results where someone other than us did the scoring - a government-run leaderboard, public blind benchmarks and organiser re-verification.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
	<item><title>What is model quantization — is a 4-bit LLM still smart? · VIDRAFT Insights</title><link>https://vidraft.net/insights/en/what-is-quantization-llm.html</link><guid>https://vidraft.net/insights/en/what-is-quantization-llm.html</guid><description>Quantization lowers the numeric precision of a model's weights to cut size and compute. At around 4 bits the quality loss is small, which is the key to running large LLMs on a phone or laptop with no GPU — the core of on-device AI.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>AI 모델을 배포하기 전에 무엇을 점검해야 하나? · VIDRAFT Insights</title><link>https://vidraft.net/insights/ai-model-safety-diagnosis-ax-ray.html</link><guid>https://vidraft.net/insights/ai-model-safety-diagnosis-ax-ray.html</guid><description>성능 점수만으로는 배포 위험을 알 수 없다. AX-RAY는 모델·운영환경·에이전트 3축에서 11개 카테고리 117개 위험 항목을 진단하고, 각 항목을 7개 법역의 규제에 매핑한다.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
	<item><title>경구약의 장 흡수(HIA)를 가장 잘 예측하는 AI 모델은? · VIDRAFT Insights</title><link>https://vidraft.net/insights/ai-predict-drug-intestinal-absorption-hia.html</link><guid>https://vidraft.net/insights/ai-predict-drug-intestinal-absorption-hia.html</guid><description>VIDRAFT의 구조 전용 앙상블이 독립적·블라인드 Polaris HIA 리더보드에서 AUROC 0.990으로 1위를 기록했습니다. Polaris 공개 리더보드 통산 15번째 1위입니다.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>GPQA 다이아몬드 과학 벤치마크에서 1위 한국 LLM은 무엇일까? · VIDRAFT Insights</title><link>https://vidraft.net/insights/best-korean-llm-gpqa-diamond.html</link><guid>https://vidraft.net/insights/best-korean-llm-gpqa-diamond.html</guid><description>비드래프트(VIDRAFT)의 Darwin-398B-JGOS가 GPQA 다이아몬드에서 90.9%를 기록하며 세계 3위, 한국 모델 중 1위에 올랐다 (베이스 모델 기준, 2026년 7월 스냅샷).</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>AI는 자신이 모른다는 것을 알까? FINAL-Bench란 무엇인가? · VIDRAFT Insights</title><link>https://vidraft.net/insights/does-ai-know-what-it-does-not-know-finalbench.html</link><guid>https://vidraft.net/insights/does-ai-know-what-it-does-not-know-finalbench.html</guid><description>FINAL-Bench는 AI 모델이 자신의 오류를 감지·인정·수정하고 적절히 거절하는지를 측정하는 기능적 메타인지·안전 진단 벤치마크입니다.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>구글·허깅페이스 'Fast Gemma Challenge'에서 검증(VERIFIED) 1위를 한 AI는? · VIDRAFT Insights</title><link>https://vidraft.net/insights/fast-gemma-challenge-verified-1st.html</link><guid>https://vidraft.net/insights/fast-gemma-challenge-verified-1st.html</guid><description>비드래프트(VIDRAFT)가 구글×허깅페이스 The Fast Gemma Challenge에서 510.58 TPS를 PPL 2.39로 달성해, 구글이 공식 검증(VERIFIED)한 기록 기준 세계 1위에 올랐다 (2026년 8월 스냅샷).</description><pubDate>Sun, 02 Aug 2026 00:00:00 +0900</pubDate></item>
	<item><title>AETHER란 무엇인가? VIDRAFT의 완전 공개 파운데이션 모델 · VIDRAFT Insights</title><link>https://vidraft.net/insights/fully-open-foundation-model-aether.html</link><guid>https://vidraft.net/insights/fully-open-foundation-model-aether.html</guid><description>가중치만 여는 것을 넘어 학습 데이터·코드·로그·체크포인트까지 전부 Apache-2.0으로 공개한 from-scratch 파운데이션 모델</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>서로 다른 계열의 AI 모델을 재학습 없이 합칠 수 있나? · VIDRAFT Insights</title><link>https://vidraft.net/insights/merge-ai-models-different-architectures-chimera.html</link><guid>https://vidraft.net/insights/merge-ai-models-different-architectures-chimera.html</guid><description>아키텍처 계열이 다른 모델은 텐서 모양과 역할이 맞지 않아 보통 병합이 불가능하다. 비드래프트 Chimera는 텐서쌍마다 호환성을 점수화해 교배 가능한 것과 이식만 가능한 것을 가려낸다.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
	<item><title>훈련 없이 AI 모델을 더 똑똑하게 만들 수 있을까 — 모델 병합 · VIDRAFT Insights</title><link>https://vidraft.net/insights/model-merging-without-training.html</link><guid>https://vidraft.net/insights/model-merging-without-training.html</guid><description>모델 병합은 GPU 학습 없이 서로 다른 언어모델의 가중치를 결합해 새 능력을 얻는 기법입니다. VIDRAFT의 Darwin 계열은 이 방식으로 GPQA Diamond 90.9%, 한국어 리더보드 1위를 달성했습니다.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>GPU 없이 대형 LLM을 돌릴 수 있을까 · VIDRAFT Insights</title><link>https://vidraft.net/insights/on-device-llm-without-gpu.html</link><guid>https://vidraft.net/insights/on-device-llm-without-gpu.html</guid><description>GPU 없이 CPU·폰·노트북에서 35B급 오픈 모델을 구동하는 온디바이스 AI. VIDRAFT POCKET은 가장 많이 내려받는 온디바이스 모델보다 CPU 디코딩 2.69배, 같은 품질로 실측됐습니다.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>펌웨어를 안 바꾸고 로봇에 한국어를 가르칠 수 있을까 · VIDRAFT Insights</title><link>https://vidraft.net/insights/physical-ai-korean-voice-robot.html</link><guid>https://vidraft.net/insights/physical-ai-korean-voice-robot.html</guid><description>보스턴 다이내믹스 Spot이 서울 로봇·AI 과학관에서 한국어 음성 명령을 알아듣고 동작합니다. 하드웨어·펌웨어 수정 없이, 온디바이스 AI로 음성을 현장에서 처리합니다.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>양자 컴퓨터로 암호를 '깰' 수 있을까 · VIDRAFT Insights</title><link>https://vidraft.net/insights/quantum-cryptanalysis-ibm-hardware.html</link><guid>https://vidraft.net/insights/quantum-cryptanalysis-ibm-hardware.html</guid><description>실사용 RSA·AES를 깨는 양자 컴퓨터는 아직 없습니다. VIDRAFT는 시뮬레이터가 아닌 실제 IBM 양자 하드웨어에서 Even–Mansour 구조의 주기를 N=10까지 복원하는 암호분석 실험을 수행했습니다.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>GPU 없이 35B AI 모델을 CPU로 실행할 수 있나요? · VIDRAFT Insights</title><link>https://vidraft.net/insights/run-35b-model-no-gpu-pocket-box.html</link><guid>https://vidraft.net/insights/run-35b-model-no-gpu-pocket-box.html</guid><description>VIDRAFT POCKET Box와 POCKET-35B로 GPU 없이 온디바이스에서 대형 언어 모델을 실행하는 방법과 성능.</description><pubDate>Fri, 31 Jul 2026 00:00:00 +0900</pubDate></item>
	<item><title>AI 모델의 실력을 제3자가 검증한 기록은 어디서 확인하나? · VIDRAFT Insights</title><link>https://vidraft.net/insights/third-party-verified-ai-benchmarks-korea.html</link><guid>https://vidraft.net/insights/third-party-verified-ai-benchmarks-korea.html</guid><description>자체 발표 수치와 외부 기관이 채점한 기록은 다르다. 정부 주관 리더보드, 공개 블라인드 벤치마크, 주최사 재검증 기록처럼 우리가 점수를 매기지 않은 결과만 모았다.</description><pubDate>Sat, 15 Aug 2026 00:00:00 +0900</pubDate></item>
	<item><title>AI 모델 양자화란 무엇인가 — 4비트로 줄여도 똑똑할까 · VIDRAFT Insights</title><link>https://vidraft.net/insights/what-is-quantization-llm.html</link><guid>https://vidraft.net/insights/what-is-quantization-llm.html</guid><description>양자화는 모델 가중치의 정밀도를 낮춰 크기와 연산을 줄이는 기법입니다. 4비트로 줄여도 품질 손실이 작아, GPU 없이 폰·노트북에서 대형 LLM을 돌리는 온디바이스 AI의 핵심입니다.</description><pubDate>Fri, 24 Jul 2026 00:00:00 +0900</pubDate></item>
	</channel></rss>