English(EN) Who Pays for the Knowledge Models Copy? Model Collapse as Rogers' Paradox

模型坍塌：罗杰斯悖论解释生成式AI的知识流失

作者 PulseAugur 编辑部 · [1 个来源] · 2026-11-05 05:00

一篇新论文探讨了“模型坍塌”现象，即在先前模型输出上训练的生成模型会随着代际的演进而逐渐丢失知识。研究人员提出，可以通过罗杰斯悖论来理解这一问题，该悖论表明，仅依赖复制的数据会导致停滞，除非不断引入新信息。论文通过分析语言模型来检验这一理论，发现它们在合成数据上训练时保留的稀有知识较少。研究还表明，当模型在新鲜数据上训练或从固定的预训练权重重新启动时，它们对世界变化的适应性更强，并且用户对模型输出的验证对于打破坍塌循环至关重要。 AI

影响强调了AI模型能力在代际之间可能出现的退化，并强调了新数据来源的必要性。

排序理由学术论文，讨论了应用于AI模型的理论概念。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 26 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 学术论文，讨论了应用于AI模型的理论概念。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, model release; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Nam H. Le · 2026-11-05 05:00

知识模型复制的成本由谁承担？罗杰斯悖论引发的模型崩溃

arXiv:2611.06022v1 Announce Type: cross Abstract: Generative models trained on the output of earlier models lose knowledge across generations, rare knowledge first. The remedies that work all bring in information from outside the loop, and the theory of model collapse treats that…

本文如何被排名

报道来源 [1]

知识模型复制的成本由谁承担？罗杰斯悖论引发的模型崩溃

相关实体

相关话题