📊 Статистика дайджестов
Всего дайджестов: 34022 Добавлено сегодня: 82
Последнее обновление: сегодня
📄 On the Fundamental Limits of LLMs at Scale
2025-11-19Авторы:
Muhammad Ahmed Mohsin, Muhammad Umer, Ahsan Bilal, Zeeshan Memon, Muhammad Ibtsaam Qadir, Sagnik Bhattacharya, Hassan Rizwan, Abhiram R. Gorle, Maahe Zehra Kazmi, Ayesha Mohsin, Muhammad Usman Rafique, Zihao He, Pulkit Mehta, Muhammad Ali Jamshed, John M. Cioffi
Саммари на русском не найдено
Доступные поля: ['id', 'arxiv_id', 'title', 'authors', 'abstract', 'summary_ru', 'categories', 'published_date', 'created_at']
Доступные поля: ['id', 'arxiv_id', 'title', 'authors', 'abstract', 'summary_ru', 'categories', 'published_date', 'created_at']
Annotation:
Large Language Models (LLMs) have benefited enormously from scaling, yet these gains are bounded by five fundamental limitations: (1) hallucination, (2) context compression, (3) reasoning degradation, (4) retrieval fragility, and (5) multimodal misalignment. While existing surveys describe these phenomena empirically, they lack a rigorous theoretical synthesis connecting them to the foundational limits of computation, information, and learning. This work closes that gap by presenting a unified, ...