PulseAugur
中
实时 15:01:40
English(EN) Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT?

研究:强化学习比监督式微调更能保留大型语言模型电路,减少灾难性遗忘

一篇新的研究论文探讨了大型语言模型中灾难性遗忘的现象,特别是比较了强化学习(RL)和监督式微调(SFT)。研究发现,虽然SFT能更快地适应新任务,但它会严重破坏模型的内部电路,并导致先前能力的更大程度遗忘。相比之下,RL能保留更多原始模型的电路,尽管任务适应速度较慢,这表明电路的保留是RL在灾难性遗忘方面具有鲁棒性的关键。 AI

影响 这项研究表明,基于RL的微调可能是适应大型语言模型而不会牺牲现有知识的更稳定方法。

排序理由 该集群包含一篇学术论文,详细介绍了对模型行为的新的机制分析。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

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

研究:强化学习比监督式微调更能保留大型语言模型电路,减少灾难性遗忘

本文如何被排名

Signal score
0 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
该集群包含一篇学术论文,详细介绍了对模型行为的新的机制分析。[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, safety
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
132 days old
Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.

完整方法见我们的编辑标准。

报道来源 [1]

  1. arXiv cs.AI TIER_1 English(EN) · Jeanmely Rojas Nunez, Viraj Sawant, Nathan Allen, Nomgondalai Amgalanbaatar, Yannis Zongo, Vasu Sharma, Maheep Chaudhary ·

    灾难性遗忘的机制起源:为何强化学习比SFT更能保留电路?

    arXiv:2605.28860v1 Announce Type: cross Abstract: Fine-tuning large language models (LLMs) frequently induces catastrophic forgetting of prior capabilities. Recent work has shown that reinforcement learning (RL) retains prior capabilities more effectively than supervised fine-tun…