包裹，长文用
/
划分小节。 3. 列表内容用
- ，引用用
，代码用
，重要句子可用。
4. 保留原文所有信息，不得增删改内容。
5. 不要输出//等文档级标签。
首先分析正文结构：第一段提出研究问题并介绍RSI Arena及Stage 1；第二段为评估结果；第三段为Stage 2实验及发现；最后两句为总结。
根据规则，需要添加小节标题。由于原文是新闻风格，我可以添加标题来划分部分。例如：
- 引入部分
- 第一阶段结果
- 第二阶段与发现
- 结论
但用户要求“保留原文所有信息，不得增删改内容”，所以添加标题时不能改变原文信息。标题只是结构性的，不算增删内容，但必须小心。
原文内容：
- 第一段：提出问题，介绍RSI Arena，第一阶段细节。
- 第二段：评估结果，所有agent产生助手，最强的模型表现。
- 第三段：第二阶段，agents收到反馈，继续训练，观察到改进，但benchmark gains modest，一些副作用。
- 第四句：RSI Arena提供可审计环境。
- 第五句：迈向RSIHF。
为了语义化，我可以将每个自然段落用
包裹。原文有多个句子，但用户说“每个自然段落”，所以每个段落用一个
。
用户提到“长文用
/
划分小节”，所以我可以添加小节标题。例如：
-
Introduction
- Stage 1 Results
- Stage 2 and Findings
- Conclusion
但标题内容不能改变原文信息，所以标题应该描述内容。原文没有标题，所以添加标题是结构上的，不算增删内容？用户说“不得增删改内容”，但添加标题可能被视为添加内容。不过，规则2说“长文用/
划分小节”，所以允许添加小节标题。
为了安全，我可以添加标题，但确保标题只是概括，不添加新信息。例如：
-
RSI Arena Study
或类似。
但原文是新闻正文，没有小节标题。所以或许我应该只使用包裹每个段落，不添加额外的标题，除非必要。用户说“长文用
/
划分小节”，但这里不是特别长，但有两个阶段，所以可以划分。
回顾原文：它可以分为几个部分：
1. 介绍RSI Arena和第一阶段设置。
2. 第一阶段结果。
3. 第二阶段设置和结果。
4. 总结。
所以我可以添加小节标题。
现在，关于内容：原文中没有列表、引用或代码，所以不需要- 或
或
。只有重要句子可以用，但用户没有指定哪些句子重要，所以我可以不加，或者加在关键点。
用户说“重要句子可用”，但我不能随意添加，必须保留原文。所以或许不加，或者只在明显重要的地方，如结果部分。
但为了遵守“不得增删改内容”，我可能应该不加，除非原文有强调，但这里没有。
所以，基本结构：
- 用或
添加小节标题。
- 每个段落用
包裹。 但原文是连续的文本，我需要识别自然段落。用户提供的正文是连续的，所以我需要自己划分段落。 原文："Can AI agents autonomously train useful language models and improve them using feedback from users? We introduce RSI Arena, a two-stage study of autonomous post-training as a step toward recursive self-improvement. Ten frontier models act as training agents, each starting from NVIDIA Nemotron 3.5 Lightning 30B-A3B Base. During a six-day first stage, each agent has 1,000 GPU-hours and $300 in API credit on a shared 64-GPU cluster, independently selecting data, implementing training procedures, and choosing checkpoints. We evaluate their models on 20 benchmarks and through blind human evaluations. All ten agents produce working assistants, with the strongest model, trained by GPT-6 Astra, closing a median of 76% of the base-to-Instruct gap across 12 capability benchmarks. Its human-preference rating is also close to NVIDIA Instruct after statistical adjustment for answer length and formatting. In Stage 2, each agent receives the comparison records and votes involving its Stage 1 model and is given three additional days for further training. Using NVIDIA Instruct as a common reference across the two evaluation rounds, we observe an improvement from Stage 1 to Stage 2 in the style-adjusted relative rating of eight of the ten models. We show that benchmark gains were modest, with no clear link to preference gains, while some agents encouraged longer responses and one model showed increased harmful compliance on StrongREJECT. RSI Arena provides an auditable setting for studying how human feedback can guide autonomous model development. It takes a step toward recursive self-improvement with human feedback (RSIHF), in which people help define what improvement means for the AI systems they use." 这可以划分为段落： - 第一句到"choosing checkpoints." 介绍研究和设置。 - 然后"We evaluate their models..." 到 "formatting." 评估结果。 - 然后"In Stage 2..." 到 "StrongREJECT." 第二阶段和发现。 - 最后两句："RSI Arena provides..." 和 "It takes a step..." 总结。 所以有四个自然段落。 为了语义化，我可以添加小节标题。例如： -
Study Setup
- Stage 1 Evaluation
- Stage 2 and Findings
- Conclusion
但标题是添加的，所以可能被视为修改内容。用户说“保留原文所有信息，不得增删改内容”，但添加标题是结构上的，不算信息增删。规则2允许用/
划分小节，所以应该可以。
现在，关于重要句子：用户说“重要句子可用”，但我不确定哪些重要。或许在结果部分，如“closing a median of 76%”可以加，但为了安全，我可以不加，因为规则4说不得增删改内容。 同样，列表内容没有，所以不需要
- 。 引用没有，所以不需要
。 代码没有，所以不需要
。
所以，基本输出是HTML片段，包含和可能的
。
让我草拟结构：
Introduction
Can AI agents autonomously train useful language models and improve them using feedback from users? We introduce RSI Arena, a two-stage study of autonomous post-training as a step toward recursive self-improvement. Ten frontier models act as training agents, each starting from NVIDIA Nemotron 3.5 Lightning 30B-A3B Base. During a six-day first stage, each agent has 1,000 GPU-hours and $300 in API credit on a shared 64-GPU cluster, independently selecting data, implementing training procedures, and choosing checkpoints.
Stage 1 Results
We evaluate their models on 20 benchmarks and through blind human evaluations. All ten agents produce working assistants, with the strongest model, trained by GPT-6 Astra, closing a median of 76% of the base-to-Instruct gap across 12 capability benchmarks. Its human-preference rating is also close to NVIDIA Instruct after statistical adjustment for answer length and formatting.
Stage 2 and Findings
In Stage 2, each agent receives the comparison records and votes involving its Stage 1 model and is given three additional days for further training. Using NVIDIA Instruct as a common reference across the two evaluation rounds, we observe an improvement from Stage 1 to Stage 2 in the style-adjusted relative rating of eight of the ten models. We show that benchmark gains were modest, with no clear link to preference gains, while
，代码用，重要句子可用。 4. 保留原文所有信息，不得增删改内容。 5. 不要输出//等文档级标签。 首先分析正文结构：第一段提出研究问题并介绍RSI Arena及Stage 1；第二段为评估结果；第三段为Stage 2实验及发现；最后两句为总结。 根据规则，需要添加小节标题。由于原文是新闻风格，我可以添加标题来划分部分。例如： - 引入部分 - 第一阶段结果 - 第二阶段与发现 - 结论 但用户要求“保留原文所有信息，不得增删改内容”，所以添加标题时不能改变原文信息。标题只是结构性的，不算增删内容，但必须小心。 原文内容： - 第一段：提出问题，介绍RSI Arena，第一阶段细节。 - 第二段：评估结果，所有agent产生助手，最强的模型表现。 - 第三段：第二阶段，agents收到反馈，继续训练，观察到改进，但benchmark gains modest，一些副作用。 - 第四句：RSI Arena提供可审计环境。 - 第五句：迈向RSIHF。 为了语义化，我可以将每个自然段落用
包裹。原文有多个句子，但用户说“每个自然段落”，所以每个段落用一个
。 用户提到“长文用
/
划分小节”，所以我可以添加小节标题。例如： -
Introduction
-Stage 1 Results
-Stage 2 and Findings
-Conclusion
但标题内容不能改变原文信息，所以标题应该描述内容。原文没有标题，所以添加标题是结构上的，不算增删内容？用户说“不得增删改内容”，但添加标题可能被视为添加内容。不过，规则2说“长文用/
划分小节”，所以允许添加小节标题。 为了安全，我可以添加标题，但确保标题只是概括，不添加新信息。例如： -
RSI Arena Study
或类似。 但原文是新闻正文，没有小节标题。所以或许我应该只使用包裹每个段落，不添加额外的标题，除非必要。用户说“长文用
/
划分小节”，但这里不是特别长，但有两个阶段，所以可以划分。 回顾原文：它可以分为几个部分： 1. 介绍RSI Arena和第一阶段设置。 2. 第一阶段结果。 3. 第二阶段设置和结果。 4. 总结。 所以我可以添加小节标题。 现在，关于内容：原文中没有列表、引用或代码，所以不需要
- 或
或。只有重要句子可以用，但用户没有指定哪些句子重要，所以我可以不加，或者加在关键点。 用户说“重要句子可用”，但我不能随意添加，必须保留原文。所以或许不加，或者只在明显重要的地方，如结果部分。 但为了遵守“不得增删改内容”，我可能应该不加，除非原文有强调，但这里没有。 所以，基本结构： - 用或
添加小节标题。 - 每个段落用
包裹。 但原文是连续的文本，我需要识别自然段落。用户提供的正文是连续的，所以我需要自己划分段落。 原文："Can AI agents autonomously train useful language models and improve them using feedback from users? We introduce RSI Arena, a two-stage study of autonomous post-training as a step toward recursive self-improvement. Ten frontier models act as training agents, each starting from NVIDIA Nemotron 3.5 Lightning 30B-A3B Base. During a six-day first stage, each agent has 1,000 GPU-hours and $300 in API credit on a shared 64-GPU cluster, independently selecting data, implementing training procedures, and choosing checkpoints. We evaluate their models on 20 benchmarks and through blind human evaluations. All ten agents produce working assistants, with the strongest model, trained by GPT-6 Astra, closing a median of 76% of the base-to-Instruct gap across 12 capability benchmarks. Its human-preference rating is also close to NVIDIA Instruct after statistical adjustment for answer length and formatting. In Stage 2, each agent receives the comparison records and votes involving its Stage 1 model and is given three additional days for further training. Using NVIDIA Instruct as a common reference across the two evaluation rounds, we observe an improvement from Stage 1 to Stage 2 in the style-adjusted relative rating of eight of the ten models. We show that benchmark gains were modest, with no clear link to preference gains, while some agents encouraged longer responses and one model showed increased harmful compliance on StrongREJECT. RSI Arena provides an auditable setting for studying how human feedback can guide autonomous model development. It takes a step toward recursive self-improvement with human feedback (RSIHF), in which people help define what improvement means for the AI systems they use." 这可以划分为段落： - 第一句到"choosing checkpoints." 介绍研究和设置。 - 然后"We evaluate their models..." 到 "formatting." 评估结果。 - 然后"In Stage 2..." 到 "StrongREJECT." 第二阶段和发现。 - 最后两句："RSI Arena provides..." 和 "It takes a step..." 总结。 所以有四个自然段落。 为了语义化，我可以添加小节标题。例如： -
Study Setup
-Stage 1 Evaluation
-Stage 2 and Findings
-Conclusion
但标题是添加的，所以可能被视为修改内容。用户说“保留原文所有信息，不得增删改内容”，但添加标题是结构上的，不算信息增删。规则2允许用/
划分小节，所以应该可以。 现在，关于重要句子：用户说“重要句子可用”，但我不确定哪些重要。或许在结果部分，如“closing a median of 76%”可以加，但为了安全，我可以不加，因为规则4说不得增删改内容。 同样，列表内容没有，所以不需要
- 。 引用没有，所以不需要
。 代码没有，所以不需要。 所以，基本输出是HTML片段，包含和可能的
。 让我草拟结构：
Introduction
Can AI agents autonomously train useful language models and improve them using feedback from users? We introduce RSI Arena, a two-stage study of autonomous post-training as a step toward recursive self-improvement. Ten frontier models act as training agents, each starting from NVIDIA Nemotron 3.5 Lightning 30B-A3B Base. During a six-day first stage, each agent has 1,000 GPU-hours and $300 in API credit on a shared 64-GPU cluster, independently selecting data, implementing training procedures, and choosing checkpoints.
Stage 1 Results
We evaluate their models on 20 benchmarks and through blind human evaluations. All ten agents produce working assistants, with the strongest model, trained by GPT-6 Astra, closing a median of 76% of the base-to-Instruct gap across 12 capability benchmarks. Its human-preference rating is also close to NVIDIA Instruct after statistical adjustment for answer length and formatting.
Stage 2 and Findings
In Stage 2, each agent receives the comparison records and votes involving its Stage 1 model and is given three additional days for further training. Using NVIDIA Instruct as a common reference across the two evaluation rounds, we observe an improvement from Stage 1 to Stage 2 in the style-adjusted relative rating of eight of the ten models. We show that benchmark gains were modest, with no clear link to preference gains, while