{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-07-28T06:20:51.496Z","headline":"Apple 提出 LEAD 方法，破解长程推理中的\"不可恢复瓶颈\"","description":"Apple 研究发现，大语言模型在长程执行中即使有高层策略也不稳定，极端分解会导致\"不可恢复瓶颈\"--少数\"困难\"步骤上的持续错误变得不可逆转。为此提出 Lookahead-Enhanced Atomic Decomposition（LEAD），通过引入短程未来验证与聚合来打破这一瓶颈。该方法在受控算法谜题上验证了有效性。","url":"https://www.aioga.com/news/cmrz48xjc01obroeyrxf2h5bq/","mainEntityOfPage":"https://www.aioga.com/news/cmrz48xjc01obroeyrxf2h5bq/","datePublished":"2026-07-24T00:00:00.000Z","dateModified":"2026-07-24T00:00:00.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://machinelearning.apple.com/research/lead-no-recovery-bottleneck","https://aihot.virxact.com/items/cmrz48xjc01obroeyrxf2h5bq"],"canonicalUrl":"https://www.aioga.com/news/cmrz48xjc01obroeyrxf2h5bq/","directAnswer":{"@type":"Answer","text":"Aioga 编辑摘要：Apple 研究发现，大语言模型在长程执行中即使有高层策略也不稳定，极端分解会导致\"不可恢复瓶颈\"--少数\"困难\"步骤上的持续错误变得不可逆转。 Aioga 将其归入「论文研究」方向，重点关注它对真实使用和行业竞争的影响。","url":"https://www.aioga.com/news/cmrz48xjc01obroeyrxf2h5bq/","dateCreated":"2026-07-24T00:00:00.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"machinelearning.apple.com source article","url":"https://machinelearning.apple.com/research/lead-no-recovery-bottleneck","datePublished":"2026-07-24T00:00:00.000Z","provider":{"@type":"Organization","name":"machinelearning.apple.com","url":"https://machinelearning.apple.com/research/lead-no-recovery-bottleneck"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmrz48xjc01obroeyrxf2h5bq","datePublished":"2026-07-24T00:00:00.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmrz48xjc01obroeyrxf2h5bq"}}],"aggregationSource":"Apple Machine Learning Research（RSS）","originalPublisher":{"name":"machinelearning.apple.com","url":"https://machinelearning.apple.com/research/lead-no-recovery-bottleneck"},"article":{"id":"cmrz48xjc01obroeyrxf2h5bq","slug":"cmrz48xjc01obroeyrxf2h5bq","url":"https://www.aioga.com/news/cmrz48xjc01obroeyrxf2h5bq/","title":"Apple 提出 LEAD 方法，破解长程推理中的\"不可恢复瓶颈\"","title_en":"LEAD： Breaking the No-Recovery Bottleneck in Long-Horizon Reasoning","summary":"Apple 研究发现，大语言模型在长程执行中即使有高层策略也不稳定，极端分解会导致\"不可恢复瓶颈\"--少数\"困难\"步骤上的持续错误变得不可逆转。为此提出 Lookahead-Enhanced Atomic Decomposition（LEAD），通过引入短程未来验证与聚合来打破这一瓶颈。该方法在受控算法谜题上验证了有效性。","source":"Apple Machine Learning Research（RSS）","sourceUrl":"https://machinelearning.apple.com/research/lead-no-recovery-bottleneck","aiHotUrl":"https://aihot.virxact.com/items/cmrz48xjc01obroeyrxf2h5bq","publishedAt":"2026-07-24T00:00:00.000Z","category":"论文研究","score":51,"selected":false,"articleBody":["LEAD: Breaking the No-Recovery Bottleneck in Long-Horizon Reasoning","Authors Denys Pushkin†, Emmanuel Abbé†","View publication：https://arxiv.org/abs/2603.06870","Long-horizon execution in Large Language Models (LLMs) remains unstable even when high-level strategies are provided. Evaluating on controlled algorithmic puzzles, we demonstrate that while decomposition is essential for stability, extreme decomposition creates a “no-recovery bottleneck”. We show that this bottleneck becomes critical due to highly non-uniform error distribution, where consistent errors on a few “hard” steps become irreversible. To address this, we propose Lookahead-Enhanced Atomic Decomposition (LEAD). By incorporating short-horizon future validation and aggregating overlapping rollouts, LEAD provides enough isolation to maintain stability while retaining enough local context to correct errors. This enables the o4-mini model to solve Checkers Jumping up to complexity n = 13, whereas extreme decomposition fails beyond n = 11.","Divide-or-Conquer? Which Part Should You Distill Your LLM?","October 25, 2024 research area Speech and Natural Language Processing：/research/?domain=Speech%20and%20Natural%20Language%20Processing conference EMNLP：/research/?event=EMNLP","Recent methods have demonstrated that Large Language Models (LLMs) can solve reasoning tasks better when they are encouraged to solve subtasks of the main task first. In this paper we devise a similar strategy that breaks down reasoning tasks into a problem decomposition phase and a problem solving phase and show that the strategy is able to outperform a single stage solution. Further, we hypothesize that the decomposition should be easier to…","Joint Learning of Portrait Intrinsic Decomposition and Relighting","July 28, 2021 research area Computer Vision：/research/?domain=Computer%20Vision","Inverse rendering is the problem of decomposing an image into its intrinsic components, i.e. albedo, normal and lighting. To solve this ill-posed problem from single image, state-of-the-art methods in shape from shading mostly resort to supervised training on all the components on either synthetic or real datasets. Here, we propose a new self-supervised training paradigm that 1) reduces the need for full supervision on the decomposition task and…","Our research in machine learning breaks new ground every day."],"articleImages":[{"sourceUrl":"https://mlr.cdn-apple.com/media/Discover_1440x420_2x_9c465d585e.jpg","alt":"Bottom banner","afterParagraph":9,"url":"/media/articles/cmrz48xjc01obroeyrxf2h5bq/64c2324784f2cf67.jpg"}],"mediaStatus":"ok","articleBodyZh":["LEAD：打破长远推理中的不可恢复瓶颈","作者 Denys Pushkin†, Emmanuel Abbé†","查看出版物：https://arxiv.org/abs/2603.06870","即使提供高层策略，大型语言模型（LLMs）在长远执行中仍然不稳定。在受控的算法谜题测试中，我们证明了虽然分解对于稳定性至关重要，但过度分解会产生“不可恢复瓶颈”。我们展示了，这个瓶颈之所以变得关键，是由于高度不均匀的错误分布，其中在少数“困难”步骤上的持续错误变得不可逆。为了解决这一问题，我们提出了前瞻增强原子分解（LEAD）。通过结合短期未来验证和聚合重叠回滚，LEAD 提供了足够的隔离以维持稳定，同时保留足够的局部上下文来纠正错误。这使得 o4-mini 模型能够解决跳棋问题，复杂度可达 n = 13，而极端分解在 n > 11 时失败。","分而治之？你应该蒸馏 LLM 的哪一部分？","2024 年 10 月 25 日 研究领域：语音与自然语言处理 /research/?domain=Speech%20and%20Natural%20Language%20Processing 会议：EMNLP /research/?event=EMNLP","最近的方法表明，当鼓励大型语言模型（LLMs）先解决主任务的子任务时，它们在推理任务中表现更好。在本文中，我们设计了类似的策略，将推理任务分解为问题分解阶段和问题解决阶段，并表明该策略能够优于单阶段解决方案。此外，我们假设分解阶段应更容易...","人像固有分解与重光照的联合学习","2021 年 7 月 28 日 研究领域：计算机视觉 /research/?domain=Computer%20Vision","逆向渲染是将图像分解为其固有组成部分的问题，即反照率、法线和光照。为了从单张图像解决这个病态问题，来自阴影形状的最先进方法大多依赖于在合成或真实数据集上对所有组件进行监督训练。在这里，我们提出了一种新的自监督训练范式，该范式 1) 减少了对分解任务的全面监督需求，並…","我们的机器学习研究每天都在开辟新的领域。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Aioga 编辑摘要：Apple 研究发现，大语言模型在长程执行中即使有高层策略也不稳定，极端分解会导致\"不可恢复瓶颈\"--少数\"困难\"步骤上的持续错误变得不可逆转。 Aioga 将其归入「论文研究」方向，重点关注它对真实使用和行业竞争的影响。","background":"背景分析：模型与研究类动态需要结合能力边界、开放方式、成本、可用性和真实任务表现判断，单项指标领先不等于已经形成稳定采用。","viewpoint":"Aioga 判断：这条动态更适合作为行业观察信号，当前信息足以建立线索，但不足以推导长期结论。","implications":"影响分析：对相关团队而言，短期应先核对来源、可用范围和实际成本，再判断是否值得接入或跟进。","nextStep":"后续观察：继续观察官方文档、实际可用性、价格变化、开发者反馈和竞品回应。","evidenceRefs":["title","summary","articleBody"],"confidence":"medium","status":"published","aiGenerated":false,"autoApproved":true,"generatedBy":"rule-safe-fallback","generatedAt":"2026-07-28T06:29:10.566Z","sourceHash":"08d041ad8e30d0be","validation":{"passed":true,"mode":"rule-safe-fallback","checks":["schema","length","source-attribution","no-html"]}},"tags":["论文研究","Apple Machine Learning Research（RSS）"],"translations":{"zh-CN":{"title":"Apple 提出 LEAD 方法，破解长程推理中的\"不可恢复瓶颈\"","summary":"Apple 研究发现，大语言模型在长程执行中即使有高层策略也不稳定，极端分解会导致\"不可恢复瓶颈\"--少数\"困难\"步骤上的持续错误变得不可逆转。为此提出 Lookahead-Enhanced Atomic Decomposition（LEAD），通过引入短程未来验证与聚合来打破这一瓶颈。该方法在受控算法谜题上验证了有效性。","category":"论文研究","source":"machinelearning.apple.com","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple 提出 LEAD 方法，破解长程推理中的\"不可恢复瓶颈\" - Aioga AI资讯","description":"Apple 研究发现，大语言模型在长程执行中即使有高层策略也不稳定，极端分解会导致\"不可恢复瓶颈\"--少数\"困难\"步骤上的持续错误变得不可逆转。为此提出 Lookahead-Enhanced Atomic Decomposition（LEAD），通过引入短程未来验证与聚合来打破这一瓶颈。该方法在受控算法谜题上验证了有效性。","url":"https://www.aioga.com/news/cmrz48xjc01obroeyrxf2h5bq/"},"en":{"title":"Apple proposes the LEAD method to break the 'irrecoverable bottleneck' in long-range reasoning","summary":"Apple's research found that large language models are unstable in long-horizon execution even with high-level strategies, and extreme decomposition can lead to an 'irrecoverable bottleneck'—persistent errors on a few 'difficult' steps become irreversible. To address this, Lookahead-Enhanced Atomic Decomposition (LEAD) is proposed, which breaks this bottleneck by introducing short-term future verification and aggregation. The method has been validated on controlled algorithmic puzzles.","category":"Research","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple proposes the LEAD method to break the 'irrecoverable bottleneck' in long-range reasoning - Aioga AI News","description":"Apple's research found that large language models are unstable in long-horizon execution even with high-level strategies, and extreme decomposition can lead to an 'irrecoverable bo...","url":"https://www.aioga.com/en/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:42:30.939Z"},"ja":{"title":"Apple は LEAD 方法を提出し、長距離推論における「回復不可能なボトルネック」を解明","summary":"Appleの研究によると、大規模言語モデルは長期的な実行において、高レベルの戦略があっても安定しないことがあり、極端な分解は「回復不能なボトルネック」を引き起こす--少数の「困難な」ステップでの継続的なエラーが不可逆になる。これに対して、Lookahead-Enhanced Atomic Decomposition（LEAD）が提案され、短期的な未来の検証と集約を導入することでこのボトルネックを打破する。この手法は制御されたアルゴリズムパズルで有効性が検証された。","category":"論文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple は LEAD 方法を提出し、長距離推論における「回復不可能なボトルネック」を解明 - Aioga AIニュース","description":"Appleの研究によると、大規模言語モデルは長期的な実行において、高レベルの戦略があっても安定しないことがあり、極端な分解は「回復不能なボトルネック」を引き起こす--少数の「困難な」ステップでの継続的なエラーが不可逆になる。これに対して、Lookahead-Enhanced Atomic Decomposition（LEAD）が提案され、短期的な未来の検証と...","url":"https://www.aioga.com/ja/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:42:39.883Z"},"ko":{"title":"Apple은 장거리 추론에서 '복구 불가능 병목'을 해결하기 위해 LEAD 방법을 제안했습니다","summary":"Apple 연구에 따르면, 대형 언어 모델은 장기 실행에서 고급 전략이 있더라도 안정적이지 않으며, 극단적 분해는 '복구 불가능 병목'을 초래할 수 있는데, 일부 '어려운' 단계에서 발생하는 지속적인 오류가 돌이킬 수 없게 된다. 이를 위해 Lookahead-Enhanced Atomic Decomposition(LEAD)을 제안하였으며, 단기 미래 검증과 집계를 도입하여 이 병목을 깨는 방법이다. 이 방법은 통제된 알고리즘 퍼즐에서 유효성을 검증하였다.","category":"연구","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple은 장거리 추론에서 '복구 불가능 병목'을 해결하기 위해 LEAD 방법을 제안했습니다 - Aioga AI 뉴스","description":"Apple 연구에 따르면, 대형 언어 모델은 장기 실행에서 고급 전략이 있더라도 안정적이지 않으며, 극단적 분해는 '복구 불가능 병목'을 초래할 수 있는데, 일부 '어려운' 단계에서 발생하는 지속적인 오류가 돌이킬 수 없게 된다. 이를 위해 Lookahead-Enhanced Atomic Decomposition(LEAD)...","url":"https://www.aioga.com/ko/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:43:26.529Z"},"es":{"title":"Apple propuso el método LEAD para resolver el \"cuello de botella irrecuperable\" en el razonamiento de largo alcance","summary":"La investigación de Apple encontró que los modelos de lenguaje grandes, incluso con estrategias de alto nivel, son inestables en la ejecución a largo plazo, y la descomposición extrema puede conducir a un \"cuello de botella irreversible\": los errores persistentes en unos pocos pasos \"difíciles\" se vuelven irreversibles. Para ello, se propone la Descomposición Atómica Mejorada con Anticipación (LEAD, por sus siglas en inglés), que rompe este cuello de botella mediante la introducción de validación y agregación de futuro a corto plazo. Este método ha demostrado su efectividad en rompecabezas algorítmicos controlados.","category":"Investigación","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple propuso el método LEAD para resolver el \"cuello de botella irrecuperable\" en el razonamiento de largo alcance - Aioga Noticias de IA","description":"La investigación de Apple encontró que los modelos de lenguaje grandes, incluso con estrategias de alto nivel, son inestables en la ejecución a largo plazo, y la descomposición ext...","url":"https://www.aioga.com/es/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:43:25.079Z"},"fr":{"title":"Apple a proposé la méthode LEAD pour résoudre le \"goulot d'étranglement irrécupérable\" dans le raisonnement à long terme","summary":"La recherche d'Apple a découvert que même avec une stratégie de haut niveau, les grands modèles de langage sont instables dans l'exécution à long terme, et une décomposition extrême peut entraîner un « goulot d'étranglement irréversible » — des erreurs persistantes sur quelques étapes « difficiles » deviennent irréversibles. Pour cela, ils ont proposé la décomposition atomique améliorée par anticipation (Lookahead-Enhanced Atomic Decomposition, LEAD), qui brise ce goulot d'étranglement en introduisant une vérification et une agrégation à court terme des futurs. Cette méthode a été validée pour son efficacité sur des énigmes algorithmiques contrôlées.","category":"Recherche","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple a proposé la méthode LEAD pour résoudre le \"goulot d'étranglement irrécupérable\" dans le raisonnement à long terme - Aioga Actualités IA","description":"La recherche d'Apple a découvert que même avec une stratégie de haut niveau, les grands modèles de langage sont instables dans l'exécution à long terme, et une décomposition extrêm...","url":"https://www.aioga.com/fr/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:44:11.328Z"},"de":{"title":"Apple stellt die LEAD-Methode vor, um das \"nicht wiederherstellbare Flaschenhals\" beim Langzeit-Schlussfolgern zu überwinden","summary":"Apple-Forscher haben entdeckt, dass große Sprachmodelle auch bei Vorhandensein einer übergeordneten Strategie bei langfristiger Ausführung instabil sind, und eine extreme Zerlegung kann zu einem \"nicht wiederherstellbaren Engpass\" führen – anhaltende Fehler bei wenigen \"schwierigen\" Schritten werden irreversibel. Zu diesem Zweck wird Lookahead-Enhanced Atomic Decomposition (LEAD) vorgeschlagen, das durch die Einführung einer kurzfristigen Zukunftsüberprüfung und Aggregation diesen Engpass aufbricht. Die Methode wurde bei kontrollierten algorithmischen Rätseln auf ihre Wirksamkeit überprüft.","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple stellt die LEAD-Methode vor, um das \"nicht wiederherstellbare Flaschenhals\" beim Langzeit-Schlussfolgern zu überwinden - Aioga KI-News","description":"Apple-Forscher haben entdeckt, dass große Sprachmodelle auch bei Vorhandensein einer übergeordneten Strategie bei langfristiger Ausführung instabil sind, und eine extreme Zerlegung...","url":"https://www.aioga.com/de/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:44:08.348Z"},"pt-BR":{"title":"A Apple propôs o método LEAD para resolver o 'gargalo irreversível' em raciocínio de longo alcance","summary":"A Apple descobriu que, mesmo com estratégias de alto nível, modelos de linguagem grandes não são estáveis na execução de longo prazo, e decomposições extremas podem levar a um \"gargalo irreversível\" — erros persistentes em alguns passos \"difíceis\" se tornam irreversíveis. Para isso, foi proposta a Lookahead-Enhanced Atomic Decomposition (LEAD), que quebra esse gargalo introduzindo validação e agregação de curto prazo. O método teve sua eficácia verificada em quebra-cabeças algorítmicos controlados.","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"A Apple propôs o método LEAD para resolver o 'gargalo irreversível' em raciocínio de longo alcance - Aioga Notícias de IA","description":"A Apple descobriu que, mesmo com estratégias de alto nível, modelos de linguagem grandes não são estáveis na execução de longo prazo, e decomposições extremas podem levar a um \"gar...","url":"https://www.aioga.com/pt-BR/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:44:52.250Z"},"ru":{"title":"Apple предложила метод LEAD, решающий проблему \"невосстановимого узкого места\" в долговременных рассуждениях","summary":"Исследование Apple показало, что большие языковые модели нестабильны при долгосрочном выполнении, даже при наличии высокоуровневых стратегий, и чрезмерное разложение может привести к «необратимому узкому месту» — постоянные ошибки на нескольких «сложных» шагах становятся необратимыми. В связи с этим был предложен Lookahead-Enhanced Atomic Decomposition (LEAD), который устраняет это узкое место за счет введения краткосрочной проверки будущего и агрегации. Этот метод подтвердил свою эффективность на контролируемых алгоритмических задачах.","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple предложила метод LEAD, решающий проблему \"невосстановимого узкого места\" в долговременных рассуждениях - Aioga Новости ИИ","description":"Исследование Apple показало, что большие языковые модели нестабильны при долгосрочном выполнении, даже при наличии высокоуровневых стратегий, и чрезмерное разложение может привести...","url":"https://www.aioga.com/ru/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:45:06.893Z"},"ar":{"title":"قدمت Apple طريقة LEAD، لفك شيفرة \"عنق الزجاجة الذي لا يمكن استعادته\" في الاستدلال طويل المدى","summary":"كشفت أبحاث Apple أن النماذج اللغوية الكبيرة غير مستقرة في التنفيذ طويل المدى حتى مع وجود استراتيجيات عالية المستوى، حيث يمكن أن يؤدي التحليل المتطرف إلى \"عنق زجاجة لا يمكن استرجاعه\" — الأخطاء المستمرة في بعض الخطوات \"الصعبة\" تصبح لا رجعة فيها. لذلك تم اقتراح تحليل ذري معزز بالنظر المستقبلي (LEAD)، والذي يكسر هذا الاختناق من خلال إدخال التحقق والتجميع القصير المدى للمستقبل. وقد تم التحقق من فعالية هذه الطريقة على الألغاز الخوارزمية الخاضعة للرقابة.","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"قدمت Apple طريقة LEAD، لفك شيفرة \"عنق الزجاجة الذي لا يمكن استعادته\" في الاستدلال طويل المدى - Aioga أخبار الذكاء الاصطناعي","description":"كشفت أبحاث Apple أن النماذج اللغوية الكبيرة غير مستقرة في التنفيذ طويل المدى حتى مع وجود استراتيجيات عالية المستوى، حيث يمكن أن يؤدي التحليل المتطرف إلى \"عنق زجاجة لا يمكن استرجاعه...","url":"https://www.aioga.com/ar/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:45:55.659Z"},"hi":{"title":"Apple ने LEAD विधि प्रस्तुत की, लंबी दूरी की तर्क प्रक्रिया में 'अपरिवर्तनीय बाधा' को हल करने के लिए","summary":"Apple ने पाया कि बड़े भाषा मॉडल लंबी अवधि के निष्पादन में उच्च-स्तरीय रणनीतियों के बावजूद अस्थिर होते हैं, अत्यधिक विघटन 'अवापसीयोग्य बाधा' पैदा कर सकता है — कुछ 'कठिन' चरणों में लगातार त्रुटियां अपरिवर्तनीय हो जाती हैं। इसके लिए Lookahead-Enhanced Atomic Decomposition (LEAD) प्रस्तावित किया गया है, जो इस बाधा को तोड़ने के लिए अल्पकालिक भविष्य सत्यापन और एकत्रीकरण को शामिल करता है। इस विधि की प्रभावशीलता को नियंत्रित एल्गोरिदम पहेलियों पर सत्यापित किया गया।","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple ने LEAD विधि प्रस्तुत की, लंबी दूरी की तर्क प्रक्रिया में 'अपरिवर्तनीय बाधा' को हल करने के लिए - Aioga AI समाचार","description":"Apple ने पाया कि बड़े भाषा मॉडल लंबी अवधि के निष्पादन में उच्च-स्तरीय रणनीतियों के बावजूद अस्थिर होते हैं, अत्यधिक विघटन 'अवापसीयोग्य बाधा' पैदा कर सकता है — कुछ 'कठिन' चरणों में ल...","url":"https://www.aioga.com/hi/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:46:06.021Z"},"it":{"title":"Apple ha proposto il metodo LEAD per risolvere il \"collo di bottiglia irreversibile\" nel ragionamento a lungo raggio","summary":"La ricerca di Apple ha scoperto che i grandi modelli linguistici, anche con strategie di alto livello, sono instabili nell'esecuzione a lungo termine; la decomposizione estrema può portare a un \"collo di bottiglia irreversibile\" — errori persistenti su pochi passaggi \"difficili\" diventano irreversibili. Per questo è stato proposto il Lookahead-Enhanced Atomic Decomposition (LEAD), che rompe questo collo di bottiglia introducendo la verifica e l'aggregazione del breve termine futuro. Il metodo è stato convalidato su puzzle algoritmici controllati.","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple ha proposto il metodo LEAD per risolvere il \"collo di bottiglia irreversibile\" nel ragionamento a lungo raggio - Aioga Notizie IA","description":"La ricerca di Apple ha scoperto che i grandi modelli linguistici, anche con strategie di alto livello, sono instabili nell'esecuzione a lungo termine; la decomposizione estrema può...","url":"https://www.aioga.com/it/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:46:49.354Z"},"nl":{"title":"Apple stelde de LEAD-methode voor om de 'onherstelbare knelpunten' in langetermijnredenering te doorbreken","summary":"Apple-onderzoek heeft aangetoond dat grote taalmodellen, zelfs met hogere-strategieën, instabiel zijn bij langetermijnuitvoering, en dat extreme decompositie kan leiden tot een 'onherstelbare bottleneck' – voortdurende fouten bij een klein aantal 'lastige' stappen worden onomkeerbaar. Hiervoor wordt Lookahead-Enhanced Atomic Decomposition (LEAD) voorgesteld, dat deze bottleneck doorbreekt door kortetermijn-toekomstvalidatie en aggregatie te introduceren. Deze methode is effectief gebleken bij gecontroleerde algoritmepuzzels.","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple stelde de LEAD-methode voor om de 'onherstelbare knelpunten' in langetermijnredenering te doorbreken - Aioga AI-nieuws","description":"Apple-onderzoek heeft aangetoond dat grote taalmodellen, zelfs met hogere-strategieën, instabiel zijn bij langetermijnuitvoering, en dat extreme decompositie kan leiden tot een 'on...","url":"https://www.aioga.com/nl/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:46:49.221Z"},"tr":{"title":"Apple, uzun vadeli akıl yürütmede 'geri döndürülemez darboğazı' çözmek için LEAD yöntemini önerdi","summary":"Apple’ın araştırması, büyük dil modellerinin uzun süreli yürütmelerde yüksek seviyeli stratejilere sahip olsalar bile istikrarsız olduğunu ve aşırı ayrıştırmanın “geri dönüşü olmayan tıkanıklıklara” yol açabileceğini ortaya koydu - az sayıda “zor” adımda sürekli hataların geri alınamaz hale gelmesi. Bunun için Kısa Vadeli Gelecek Doğrulaması ve Birleştirme yoluyla bu tıkanıklığı kıran Lookahead-Enhanced Atomic Decomposition (LEAD) yöntemi önerildi. Bu yöntemin kontrollü algoritma bulmacalarında etkili olduğu doğrulandı.","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple, uzun vadeli akıl yürütmede 'geri döndürülemez darboğazı' çözmek için LEAD yöntemini önerdi - Aioga AI Haberleri","description":"Apple’ın araştırması, büyük dil modellerinin uzun süreli yürütmelerde yüksek seviyeli stratejilere sahip olsalar bile istikrarsız olduğunu ve aşırı ayrıştırmanın “geri dönüşü olmay...","url":"https://www.aioga.com/tr/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:47:37.090Z"},"vi":{"title":"Apple đưa ra phương pháp LEAD, phá vỡ 'nút thắt không thể phục hồi' trong suy luận dài hạn","summary":"Apple nghiên cứu phát hiện, các mô hình ngôn ngữ lớn trong thực thi dài hạn ngay cả khi có chiến lược cấp cao cũng không ổn định, phân rã cực đoan sẽ dẫn đến \"nghẽn không thể phục hồi\" -- các lỗi kéo dài ở một số bước \"khó khăn\" trở nên không thể đảo ngược. Do đó, đề xuất Lookahead-Enhanced Atomic Decomposition (LEAD), thông qua việc giới thiệu xác minh và tổng hợp ngắn hạn trong tương lai để phá vỡ nghẽn này. Phương pháp này đã được kiểm chứng hiệu quả trên các câu đố thuật toán được kiểm soát.","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple đưa ra phương pháp LEAD, phá vỡ 'nút thắt không thể phục hồi' trong suy luận dài hạn - Tin tức AI Aioga","description":"Apple nghiên cứu phát hiện, các mô hình ngôn ngữ lớn trong thực thi dài hạn ngay cả khi có chiến lược cấp cao cũng không ổn định, phân rã cực đoan sẽ dẫn đến \"nghẽn không thể phục...","url":"https://www.aioga.com/vi/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:47:38.385Z"},"id":{"title":"Apple mengusulkan metode LEAD, memecahkan 'bottleneck yang tidak dapat dipulihkan' dalam penalaran jangka panjang","summary":"Penelitian Apple menemukan bahwa model bahasa besar tidak stabil dalam eksekusi jangka panjang bahkan dengan strategi tingkat tinggi, dan dekomposisi ekstrem dapat menyebabkan \"titik buntu yang tidak dapat dipulihkan\"—kesalahan yang terus-menerus pada sejumlah kecil langkah \"sulit\" menjadi tidak dapat diubah. Untuk ini, diperkenalkan Lookahead-Enhanced Atomic Decomposition (LEAD), yang memecahkan titik buntu ini dengan memperkenalkan verifikasi dan agregasi masa depan jangka pendek. Metode ini telah diverifikasi efektif pada teka-teki algoritma yang terkontrol.","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple mengusulkan metode LEAD, memecahkan 'bottleneck yang tidak dapat dipulihkan' dalam penalaran jangka panjang - Berita AI Aioga","description":"Penelitian Apple menemukan bahwa model bahasa besar tidak stabil dalam eksekusi jangka panjang bahkan dengan strategi tingkat tinggi, dan dekomposisi ekstrem dapat menyebabkan \"tit...","url":"https://www.aioga.com/id/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:48:21.989Z"},"th":{"title":"Apple ได้นำเสนอวิธีการ LEAD สำหรับแก้ปัญหา 'คอขวดที่ไม่สามารถกู้คืนได้' ในการให้เหตุผลระยะยาว","summary":"การวิจัยของ Apple พบว่า แม้ว่าโมเดลภาษาขนาดใหญ่อาจมีกลยุทธ์ระดับสูง แต่ก็ยังไม่เสถียรในการทำงานระยะยาว การแยกอย่างสุดขั้วสามารถทำให้เกิด \"คอขวดที่ไม่สามารถกู้คืนได้\" — ข้อผิดพลาดที่ต่อเนื่องในขั้นตอนที่ 'ยาก' เพียงไม่กี่ขั้นตอนกลายเป็นไม่สามารถย้อนกลับได้ เพื่อแก้ปัญหานี้ จึงได้เสนอวิธี Lookahead-Enhanced Atomic Decomposition (LEAD) โดยการนำการตรวจสอบอนาคตระยะสั้นและการรวมเข้ามาช่วยทำลายคอขวดนี้ วิธีนี้ได้รับการตรวจสอบความมีประสิทธิภาพในปริศนาทางอัลกอริทึมที่ควบคุมได้","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple ได้นำเสนอวิธีการ LEAD สำหรับแก้ปัญหา 'คอขวดที่ไม่สามารถกู้คืนได้' ในการให้เหตุผลระยะยาว - ข่าว AI Aioga","description":"การวิจัยของ Apple พบว่า แม้ว่าโมเดลภาษาขนาดใหญ่อาจมีกลยุทธ์ระดับสูง แต่ก็ยังไม่เสถียรในการทำงานระยะยาว การแยกอย่างสุดขั้วสามารถทำให้เกิด \"คอขวดที่ไม่สามารถกู้คืนได้\" — ข้อผิดพลาดที...","url":"https://www.aioga.com/th/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:48:36.045Z"},"pl":{"title":"Apple przedstawiło metodę LEAD, przełamując \"nieodwracalną przeszkodę\" w długodystansowym rozumowaniu","summary":"Apple odkryło, że duże modele językowe są niestabilne w długoterminowym wykonywaniu, nawet przy istnieniu strategii wysokiego poziomu, a ekstremalny podział może prowadzić do „nieodwracalnego wąskiego gardła” – uporczywe błędy w nielicznych „trudnych” krokach stają się nieodwracalne. W związku z tym przedstawiono Lookahead-Enhanced Atomic Decomposition (LEAD), który poprzez wprowadzenie krótkoterminowej weryfikacji przyszłości i agregacji przełamuje to wąskie gardło. Metoda ta została zweryfikowana pod kątem skuteczności na kontrolowanych zagadkach algorytmicznych.","category":"论文研究","source":"Apple Machine Learning Research（RSS）","aggregationSource":"Apple Machine Learning Research（RSS）","pageTitle":"Apple przedstawiło metodę LEAD, przełamując \"nieodwracalną przeszkodę\" w długodystansowym rozumowaniu - Aioga Wiadomości AI","description":"Apple odkryło, że duże modele językowe są niestabilne w długoterminowym wykonywaniu, nawet przy istnieniu strategii wysokiego poziomu, a ekstremalny podział może prowadzić do „nieo...","url":"https://www.aioga.com/pl/news/cmrz48xjc01obroeyrxf2h5bq/","contentTranslated":true,"sourceHash":"63056e718baafbfd","translatedAt":"2026-07-27T00:49:20.222Z"}}}}