एआई दुनिया में - 20 मई, 2026 की सबसे महत्वपूर्ण घटनाएँ

आज का डेली एआई वर्ल्ड ब्रीफ दुनिया के प्रमुख क्षेत्रों से कृत्रिम बुद्धिमत्ता पर सबसे महत्वपूर्ण समाचार एकत्र करता है। ध्यान व्यावसायिक कार्यान्वयन, नियमन, सुरक्षा और एआई मॉडल के विकास पर केंद्रित है। यूरोप Designing escalation criteria for international AI incident response: criteria, triggers, and thresholds arXiv:2604.23183v2 Announce Type: replace-cross Abstract: AI incident reporting requirements are emerging in regulation and policy, yet no operational criteria exist for determining when a detected AI incident warrants escalation beyond national handling to international coordination.

This paper proposes an escalation framework to address this gap, intended as a common reference point across jurisdictions that enables aligned escalation while preserving flexibility in how actors respond within their own legal and policy contexts.

We review SB 53, the EU AI Act, the GPAI Code of Practice, and incident frameworks from other industries to derive eight criteria for assessing whether an incident warrants escalation, translated into a sequential flowchart with gated decision points and threshold checks.

For each criterion, we map how it interplays with these regulatory frameworks, identifying यह क्यों महत्वपूर्ण है: बाजार, नियमन और एआई उपयोगकर्ताओं पर इस जानकारी के प्रभाव को देखना उपयोगी है। स्रोत: arXiv AI (20.05.2026) Targeted consultation on the draft guidelines for the classification of high-risk artificial intelligence systems Targeted consultation on the draft guidelines for the classification of high-risk artificial intelligence systems marsrgi Tue, 05/19/2026 - 10:00 Opening: 19 May 2026 Closing: 23 June 2026 यह परामर्श दिशानिर्देशों की स्पष्टता और उदाहरणों की उपयोगिता पर प्रतिक्रिया एकत्र करने का लक्ष्य रखता है। ये दिशानिर्देश प्रदाताओं, डिप्लॉयर्स और अन्य संबंधित अभिनेताओं को यह निर्धारित करने में सहायता के लिए अभिप्रेत हैं कि क्या कोई एआई सिस्टम उच्च जोखिम श्रेणी में आता है। वे AI अधिनियम के प्रासंगिक प्रावधानों पर स्पष्टीकरण प्रदान करते हैं और व्यावहारिक उदाहरण शामिल करते हैं ताकि विभिन्न क्षेत्रों और उपयोग मामलों में वर्गीकरण का आकलन कैसे किया जाना चाहिए, इसका चित्रण किया जा सके। आप ड्राफ्ट दिशानिर्देश डाउनलोड कर सकते हैं। ड्राफ्ट दिशानिर्देशों को परामर्श करना आसान बनाने के लिए, उनकी सामग्री AI Act Single Information Platform पर भी एक उपयोगकर्ता-अनुकूल प्रारूप में प्रस्तुत की गई है। वहाँ, आप ड्राफ्ट गु सारांश और उदाहरणों तक जल्दी पहुँच सकते हैं यह क्यों महत्वपूर्ण है: बाजार, नियमन और एआई उपयोगकर्ताओं पर इस जानकारी के प्रभाव को देखना उपयोगी है। स्रोत: EU AI Act (19.05.2026) LP-Eval: कानूनी प्रस्ताव पीढ़ी की गुणवत्ता मापने के लिए रूब्रिक और डेटासेट arXiv:2605.19815v1 Announce Type: cross Abstract: कानूनी प्रस्तावों का निर्माण कानूनी तर्क और सैद्धांतिक विद्वता के केंद्र में है, फिर भी यह लीगल एनएलपी (Legal NLP) में कम जांचा गया है। यह पेपर लार्ज लैंग्वेज मॉडल्स (LLMs) का उपयोग करके यूरोपीय संघ के न्याय न्यायालय (Court of Justice of the European Union) के फैसलों से कानूनी प्रस्तावों के स्वचालित निर्माण और मूल्यांकन की जाँच करता है। हम LP-Eval प्रस्तुत करते हैं, जो कानूनी विशेषज्ञों के साथ सह-डिज़ाइन किया गया एक तीन-चरणीय मूल्यांकन रूब्रिक है जो कानूनी प्रस्ताव की गुणवत्ता को औपचारिक वैधता और सारभूत आयामों में विभाजित करता है। इस रूब्रिक का उपयोग करते हुए, हम 100 LLM-जनरेटेड कानूनी प्रस्तावों के लिए दो विशेषज्ञों के एनोटेशन का एक डेटासेट जारी करते हैं। हमारे परिणाम दिखाते हैं कि LLMs मुख्य रूप से सुगठित और उच्च गुणवत्ता वाले प्रस्ताव उत्पन्न कर सकते हैं, जबकि विशेषज्ञ मूल्यांकन यह प्रकट करता है कि हाल के प्रस्तावों की तुलना में अच्छी तरह स्थापित मामलों से प्राप्त प्रस्तावों की गुणवत्ता अधिक होती है। हम आगे LLMs का मूल्यांकनकर्ता (evaluator) के रूप में भी अध्ययन करते हैं। यह क्यों महत्वपूर्ण है: बाजार, नियमों और एआई उपयोगकर्ताओं पर इस जानकारी के प्रभाव को देखना उपयोगी होगा। स्रोत: arXiv AI (20.05.2026) एमएफसीसी का उपयोग करके संगीत वाद्य यंत्र पहचान के लिए डीप न्यूरल नेटवर्क arXiv:2105.00933v3 Announce Type: replace-cross Abstract: कुशल स्वचालित संगीत वर्गीकरण का कार्य अत्यंत महत्वपूर्ण है और यह संगीत क्षेत्र में AI के विभिन्न उन्नत अनुप्रयोगों का आधार बनाता है। वाद्य यंत्र पहचान, इसके ऑडियो के माध्यम से उपकरण की पहचान करने का कार्य है। इस ऑडियो, जिसे ध्वनि कंपन भी कहा जाता है, का उपयोग मॉडल द्वारा वाद्य यंत्र वर्गों से मिलान करने के लिए किया जाता है। इस पेपर में, हम एक कृत्रिम तंत्रिका नेटवर्क (ANN) मॉडल का उपयोग करते हैं जिसे बीस विभिन्न प्रकार के संगीत वाद्ययंत्रों पर वर्गीकरण करने के लिए प्रशिक्षित किया गया था। यहां हम ऑडियो डेटा के केवल मेल-फ़्रीक्वेंसी सेपस्ट्रल गुणांक (MFCCs) का उपयोग करते हैं। हमारा प्रस्तावित मॉडल पूरे लंदन फ़िलहारमोनिक ऑर्केस्ट्रा डेटासेट पर प्रशिक्षित होता है जिसमें चार परिवारों—वुडविंड्स, ब्रास, पर्कशन और स्ट्रिंग्स—से संबंधित वाद्य यंत्रों के बीस वर्ग शामिल हैं। प्रायोगिक परिणाम के आधार पर यह क्यों महत्वपूर्ण है: बाज़ार, नियमन और एआई उपयोगकर्ताओं पर इस जानकारी के प्रभाव का अवलोकन करना उपयोगी है। स्रोत: arXiv AI (20.05.2026) उत्तरी अमेरिका व्यक्तिगत स्वास्थ्य एआई में व्यक्तिगत स्वास्थ्य रिकॉर्ड की उपयोगिता का मूल्यांकन arXiv:2605.18937v1 Announce Type: new Abstract: Patient-managed Personal Health Records (PHRs) promises to empower patients to better understand their health; but information in the record is complex, potentially hindering insights.

In this study, we assess the potential of large language models (LLMs, Gemini 3.0 Flash) to provide helpful answers to user health queries, when provided clinical data from PHRs as context.

A total of 2,257 user queries were drawn from 3 different distributions to represent patient questions: shorter web search queries, longer questions derived from templates of chatbot conversations, and questions patients asked to their healthcare team (patient calls).

Queries were matched with de-identified PHRs (from a pool of 1,945).

Gemini responses were generated (1) without PHR context; (2) with a basic summary of demographics, conditions, and medications; (3) with fu यह क्यों महत्वपूर्ण है: इस जानकारी के बाजार, नियमों और एआई उपयोगकर्ताओं पर प्रभाव का अवलोकन करना उपयोगी है। स्रोत: arXiv AI (20.05.2026) OEP: Poisoning Self-Evolving LLM Agents via Locally Correct but Non-Transferable Experiences arXiv:2605.18930v1 Announce Type: cross Abstract: Memory-augmented large language model (LLM) agents use iterative reflection and self-evolution to solve complex tasks, but these mechanisms introduce security risks.