从人类反馈到大模型自我进化，看最新研究成果如何用反馈驱动LLM性能跃升。
arXiv:2607.11267v1 Announce Type: cross Abstract: In the rapidly evolving landscape of information retrieval systems, the ability to adapt and improve…
让计算机代理从失败中自我进化，推理时实时改进，打破高质量轨迹数据瓶颈。
arXiv:2606.31270v1 Announce Type: cross Abstract: Computer-use agents, which leverage multimodal large language models (MLLMs) to operate computers an…
揭示大语言模型在线自我改进对齐的收敛性理论，为LLM安全与性能优化提供数学基础，UAI 2026收录。
arXiv:2606.31524v1 Announce Type: cross Abstract: The Self-Improving Alignment (SAIL) algorithm addresses distribution shift by reducing a bilevel for…
提出一种开放式自我改进推理器OpenSIR，通过自博弈突破标注数据限制，有望超越人类水平推理。
arXiv:2511.00602v3 Announce Type: replace Abstract: Recent advances in large language model (LLM) reasoning through reinforcement learning rely on ann…
AI编码代理自主指导机器人训练，ENPIRE系统实现99%操作成功率，突破性自我改进策略
NVIDIA’s self-improvement program for robots enlists teams of AI coding agents.
LLM智能体训练新突破：用回顾性进度感知自我改进，弥补元认知缺口，提升长程任务效果
arXiv:2606.14302v1 Announce Type: new Abstract: LLM-based agents trained with reinforcement learning optimize step-wise action prediction but lack met…
AI执行力两年内从1.5小时翻至12小时，但人类在战略方向上仍不可替代
IT之家 6 月 10 日消息，Anthropic 于 6 月 5 日发布博文， 指出人工智能（AI）已开始加速 AI 本身研发，正迈步走向递归自我改进。 IT之家注：递归自我改进（recursive self-improvement）是指 AI 系统能够自主设计、训练、评估并迭代自己的下一代版本，…
Anthropic警告Claude AI正以超预期速度自我进化，现已有80%以上的代码由AI自身编写，人类控制力面临挑战。
Article URL: https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropic-says-claude-now-writes-more-than-80-percent-of-its-merged-co…
自改进的编码环境，结合LLM智能交互，支持Web端免费使用，开源且无需注册。
A coding environment designed to be "recursively self improving".... but a whole lot more. Uses web based chatbots to save tons of money while being w…
LLM智能体自我改进生成安全关键驾驶场景，实现帕累托最优进化。
arXiv:2606.03678v1 Announce Type: new Abstract: Generating safety-critical scenarios is essential for validating and improving autonomous driving syst…
探讨AI能否被信任去构建更强大的自身，直面技术伦理与安全挑战的深层拷问。
Article URL: https://www.ft.com/content/7cc7800f-18ed-47d8-9539-221ae3e16182 Comments URL: https://news.ycombinator.com/item?id=48380643 Points: 1 # C…
一个AI系统用8天自我构建完成，每次提交都标记模型作者，人类只需下发任务后喝杯咖啡回来刷新GitHub。
Article URL: https://composio.dev/blog/the-self-improving-ai-system-that-built-itself Comments URL: https://news.ycombinator.com/item?id=48370443 Poin…
为编程代理赋予连续记忆与自我进化能力，早期但核心完备，兼容Claude Code。
Article URL: https://github.com/kurikomi-labs/komi-learn Comments URL: https://news.ycombinator.com/item?id=48343216 Points: 10 # Comments: 1
LLM自我改进新方法！门控机制确保新增技能不破坏已有能力，破解程序性知识遗忘难题。
arXiv:2605.29668v1 Announce Type: new Abstract: LLM agents acting in structured environments fail in operational rather than conversational ways, and …
RSI（递归自我改进）正取代AGI成为AI新焦点，开源项目已在GPT-2模型上取得初步进展。
A new crop of AI labs are focused on recursive self-improvement — but the goal is proving elusive.
OpenAI联合Thrive和Crete打造自我进化税务AI，用Codex实现自动化报税与准确率跃升
See how OpenAI, Thrive, and Crete built a self-improving tax agent with Codex, automating filings, improving accuracy, and accelerating workflows.
用数学框架解释为何单一AI终将胜出，结合Yudkowsky与Omohundro理论推导关键定理并验证假设。
Article URL: https://github.com/ninjahawk/singleton-attractor Comments URL: https://news.ycombinator.com/item?id=48249270 Points: 1 # Comments: 0
OpenAI 开出最高 44.5 万美元年薪，招募研究 AI 自我改进风险的安全研究员，要求有品位、有策略性。
IT之家 5 月 23 日消息，据《商业内幕》23 日（今天）下午报道，OpenAI 发布了安全研究员的招聘信息，研究当 AI 能够训练出自己的“更强版本”时可能出现的问题。 职位聚合网站显示，这则招聘信息面向 OpenAI Preparedness 安全团队，本月上线。该岗位年薪高达 29.5 万…
提出单向策略优化方法，让大模型在无外反馈下自我进化，提升推理与对齐能力。
arXiv:2605.22156v1 Announce Type: cross Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has become a promising paradigm for scaling re…
联邦学习下用合成表格数据自我优化大模型，Concordia框架有望提升隐私与效率。
arXiv:2605.09855v2 Announce Type: replace Abstract: Federated learning (FL) enables training large language models (LLMs) without sharing raw data, bu…