{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-09-28T06:03:00.468Z","headline":"Google 与 DeepMind 提出 Dream-RSI，让 AI 智能体通过回放搜索历史改进策略","description":"Google 与 DeepMind 研究人员提出 Dream-RSI 方法，通过回放已完成的搜索记录离线测试数千种替代策略，只优化搜索策略而不改动底层模型。","url":"https://www.aioga.com/news/cmu8arj6g1b6erogrovd0pigf/","mainEntityOfPage":"https://www.aioga.com/news/cmu8arj6g1b6erogrovd0pigf/","datePublished":"2026-09-19T11:08:06.000Z","dateModified":"2026-09-19T11:08:06.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://the-decoder.com/google-deepminds-dream-rsi-helps-ai-agents-improve-by-dreaming-about-past-attempts","https://aihot.news/items/cmu8arj6g1b6erogrovd0pigf"],"canonicalUrl":"https://www.aioga.com/news/cmu8arj6g1b6erogrovd0pigf/","directAnswer":{"@type":"Answer","text":"Google 与 DeepMind 研究人员提出 Dream-RSI，通过回放已完成搜索的记录，离线测试替代搜索策略；该方法只调整智能体的搜索策略，不改动底层 AI 模型。","url":"https://www.aioga.com/news/cmu8arj6g1b6erogrovd0pigf/","dateCreated":"2026-09-19T11:08:06.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"the-decoder.com source article","url":"https://the-decoder.com/google-deepminds-dream-rsi-helps-ai-agents-improve-by-dreaming-about-past-attempts","datePublished":"2026-09-19T11:08:06.000Z","provider":{"@type":"Organization","name":"the-decoder.com","url":"https://the-decoder.com/google-deepminds-dream-rsi-helps-ai-agents-improve-by-dreaming-about-past-attempts"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.news/items/cmu8arj6g1b6erogrovd0pigf","datePublished":"2026-09-19T11:08:06.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.news/items/cmu8arj6g1b6erogrovd0pigf"}}],"aggregationSource":"The Decoder：AI News（RSS）","originalPublisher":{"name":"the-decoder.com","url":"https://the-decoder.com/google-deepminds-dream-rsi-helps-ai-agents-improve-by-dreaming-about-past-attempts"},"geoDeepAnswer":null,"article":{"id":"cmu8arj6g1b6erogrovd0pigf","slug":"cmu8arj6g1b6erogrovd0pigf","url":"https://www.aioga.com/news/cmu8arj6g1b6erogrovd0pigf/","title":"Google 与 DeepMind 提出 Dream-RSI，让 AI 智能体通过回放搜索历史改进策略","title_en":"","summary":"Google 与 DeepMind 研究人员提出 Dream-RSI 方法，通过回放已完成的搜索记录离线测试数千种替代策略，只优化搜索策略而不改动底层模型。","source":"The Decoder：AI News（RSS）","sourceUrl":"https://the-decoder.com/google-deepminds-dream-rsi-helps-ai-agents-improve-by-dreaming-about-past-attempts","aiHotUrl":"https://aihot.news/items/cmu8arj6g1b6erogrovd0pigf","publishedAt":"2026-09-19T11:08:06.000Z","category":"行业动态","score":58,"selected":false,"articleBody":["Researchers at Google and Deepmind have developed a method that helps AI agents tackle difficult search tasks more efficiently. It uses past search runs to test new strategies without repeating costly computations.","Self-improving AI agents are supposed to one day discover new algorithms, solutions to math problems, or faster code on their own. They follow the same basic process of proposing a solution, evaluating the result, learning from it, and trying again. Over thousands of attempts, they gradually work toward a good result.","For complex tasks, the search space can grow enormous. The agent must constantly decide which promising approaches to pursue, which to try in parallel, and which to abandon. This process, called exploration, can determine whether the search succeeds or wastes compute chasing the wrong ideas.","A research team from Google and Deepmind has introduced \"Dream-RSI：https://github.com/zhengkid/Dream-RSI\" to improve those decisions. The method changes how the agent searches, not the underlying AI model.","Existing approaches generally handle exploration in two ways. A fixed search strategy can't learn from experience, so the agent may repeatedly hit the same dead ends. Adapting the strategy during a search avoids that rigidity but comes at a cost. It takes many attempts to find out whether a strategy works, and testing countless alternatives would mean repeating long, expensive runs.","The researchers propose reusing data from a completed search to test alternative strategies within the space the agent has already explored. The agent records its attempts and their results as it searches, providing the data needed to replay those decisions later.","The researchers compare this to finding your way through an unfamiliar area. On your first visit, you hit dead ends, double back, and struggle to find a route. Once you have a mental map, though, you can plan another route without visiting every spot again.","Dream-RSI applies that principle to recorded search histories. Rather than testing a new strategy in a live run, the agent runs it against stored results. This lets it check what would have happened if it had pursued other approaches first or abandoned some earlier. The system doesn't invent entirely new solutions during replay; it tests different decisions within the recorded search tree.","Because those results already exist, the agent doesn't need to generate or evaluate solutions again, avoiding the expensive computations a live run would require. That makes testing new search strategies much cheaper. The researchers call this process \"dreaming.\" The agent plays through thousands of variations and selects the best one before putting it to work in a live search.","The process repeats in a loop. After each search, the agent uses the recorded results to test better strategies, then applies the improved version to its next live run. Throughout this cycle, only the search strategy changes; the model generating the solutions remains untouched.","The researchers tested Dream-RSI with Gemini 3.1 Pro：https://the-decoder.com/google-releases-gemini-3-1-pro-with-improved-reasoning-capabilities/?cmpscreencustom=1 and Gemini 3.7 Flash：https://the-decoder.com/gemini-3-7-flash-lands-with-coding-gains-and-undercuts-its-three-week-old-predecessors-price-by-50/ on eight tasks across three areas. Each comparison used a baseline with the same starting conditions but a fixed search strategy.","One task asked the system to write the fastest possible program for a statistical calculation commonly used in genomics and finance. Dream-RSI's program ran faster than the established libraries sklearn and glmnet on all six test datasets.","With Gemini 3.1 Pro, average runtime fell from 3,587 to 2,931 milliseconds, while the number of attempts dropped from 550 to 317. Dream-RSI also outperformed a competing system called SimpleTES, which needed 51,200 runs, compared with Dream-RSI's 317 attempts.","The same pattern held for math optimization tasks and efforts to write efficient GPU kernels, with comparable or better results at much lower computational cost. On two GPU tasks, Dream-RSI matched performance while cutting the number of runs by a factor of up to 2.43. On two others, it delivered up to 2.09 times the performance within the same budget.","In a follow-up analysis, the researchers tested another way to use search histories. Instead of replaying them to test strategies, they condensed them into instructions telling the agent where to search.","On one GPU task, the version with these instructions performed worse than the version without them. The researchers suggest that overly specific directions can narrow the search space too much, preventing the agent from exploring a broader range of approaches.","The same analysis showed how the learned strategy adjusted its effort. As performance improved, it initially reduced the number of attempts. When progress stalled, it increased the search effort again, which coincided with further gains. The researchers have shared code and more details on GitHub：https://github.com/zhengkid/Dream-RSI.","Recursive self-improvement has drawn growing attention lately. Developments in this field are part of why Anthropic CEO Dario Amodei recently warned about the pace of AI research：https://the-decoder.com/ex-deepmind-vp-vinyals-says-ai-self-improvement-is-coming-but-wont-trigger-an-intelligence-explosion/.","Google Deepmind introduced AlphaEvolve：https://the-decoder.com/alphaevolve-is-google-deepminds-new-ai-system-that-autonomously-creates-better-algorithms/ in 2025, using the same basic principle. Gemini Flash generates code proposals, Gemini Pro analyzes them, and an evolutionary algorithm selects the best versions. Dream-RSI works one level above that process by optimizing the search strategy itself.","AutoTTS：https://the-decoder.com/researchers-let-claude-code-discover-ai-scaling-algorithms-that-humans-probably-wouldnt-have-designed/?cmpscreencustom=1 takes a related approach, using a coding agent to search for algorithms in a simulated environment. These algorithms decide when a language model should start, expand, or abandon reasoning paths. The resulting methods beat manually designed methods while using less compute.","Google Research recently presented a different way to reuse past runs with WikiSkill：https://the-decoder.com/google-gives-ai-agents-their-own-wiki-so-they-can-learn-from-mistakes-and-successes/. That system records failures and successes in a wiki and turns them into reusable instructions for the agent. Dream-RSI's follow-up analysis suggests that explicit instructions like these can restrict exploration on open-ended search tasks.","Meta goes further with Hyperagents：https://the-decoder.com/metas-hyperagents-improve-at-tasks-and-improve-at-improving/?cmpscreencustom=1, allowing agents to rewrite the mechanism that controls how they improve.","Stay in the loop on AI. Clear, useful, no fluff.","Follow The Decoder for AI news, background stories and expert analyses.","The Decoder：https://the-decoder.com/"],"articleImages":[{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/09/dream-rsi-generated-image-nano-banana-pro.jpg","alt":"Image description","afterParagraph":0,"url":"/media/articles/cmu8arj6g1b6erogrovd0pigf/0ccc7404d18c3f4d.jpg"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/09/dream-rsi-02-replay-simulator-diagram.jpg","alt":"Diagram of a recorded search tree showing two alternative strategies following colored paths, each scored for quality, cost, and latency.","afterParagraph":7,"url":"/media/articles/cmu8arj6g1b6erogrovd0pigf/5cea1abf5467a896.jpg"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/09/dream-rsi-01-overview-diagram.jpg","alt":"Diagram of Dream-RSI's three-stage cycle showing live exploration, replay simulator construction, and strategy improvement through simulated searches. A close-up shows the proposal, evaluation, and feedback loop.","afterParagraph":9,"url":"/media/articles/cmu8arj6g1b6erogrovd0pigf/73428babf652639c.jpg"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/09/dream-rsi-07-exploration-behavior-chart.jpg","alt":"Line chart showing the best ConvDiv performance in rounds E0 through E8, above a bar chart showing 50 to 110 evaluated attempts per round.","afterParagraph":12,"url":"/media/articles/cmu8arj6g1b6erogrovd0pigf/1198bdbee0a5d157.jpg"},{"sourceUrl":"https://the-decoder.com/wp-content/uploads/2026/09/dream-rsi-05-kernel-results-chart.jpg","alt":"Four step charts comparing Dream-RSI with Recursive Fixed Exploration on VGG16, LayerNorm, ConvDiv, and ConvMax, plotting GPU kernel performance in inverse milliseconds against the number of generations.","afterParagraph":13,"url":"/media/articles/cmu8arj6g1b6erogrovd0pigf/c7f3c36ab7ea00b3.jpg"}],"mediaStatus":"ok","articleBodyZh":["谷歌和DeepMind的研究人员开发了一种方法，帮助AI代理更高效地处理复杂的搜索任务。该方法利用过去的搜索运行来测试新策略，而无需重复昂贵的计算。","自我改进的AI代理有望有一天能够自行发现新的算法、数学问题的解决方案或更快速的代码。它们遵循相同的基本过程：提出解决方案、评估结果、从中学习并再次尝试。经过成千上万次尝试，它们逐渐接近良好的结果。","对于复杂任务，搜索空间可能会变得极大。代理必须不断决定哪些有前景的方法值得追求，哪些可以并行尝试，以及哪些应被放弃。这个过程称为探索，它可能决定搜索是否成功，或者是否浪费计算资源去追逐错误的想法。","来自谷歌和DeepMind的一个研究团队推出了“Dream-RSI：https://github.com/zhengkid/Dream-RSI”以改进这些决策。该方法改变了智能体的搜索方式，而不是底层的人工智能模型。","现有方法通常以两种方式处理探索。固定搜索策略无法从经验中学习，因此代理可能会多次碰到相同的死胡同。在搜索过程中调整策略可以避免这种僵化，但代价不小。需要多次尝试才能确定策略是否有效，而测试无数备选方案则意味着要重复耗时且昂贵的运行。","研究人员建议重用已完成搜索的数据，在代理已经探索过的空间中测试替代策略。代理在搜索过程中记录其尝试及其结果，从而提供后续回放这些决策所需的数据。","研究人员将此方法比作在陌生区域寻找路线。第一次访问时，你会遇到死路、回头并努力寻找路径。然而，一旦你有了心理地图，就可以规划不同的路线，而无需再次访问每个地点。","Dream-RSI 将这一原理应用于记录的搜索历史。与其在实时运行中测试新策略，不如让智能体在存储的结果中运行它。这使它能够检查，如果先采用其他方法或放弃某些先前的步骤，会发生什么情况。系统不会在重放过程中完全创造新的解决方案；它是在记录的搜索树中测试不同的决策。","由于这些结果已经存在，智能体不需要再次生成或评估解决方案，从而避免了实时运行所需的昂贵计算。这使得测试新的搜索策略成本更低。研究人员称这一过程为“做梦”。智能体会演练成千上万种变体，并选择最佳方案，然后再将其应用于实时搜索。","这一过程以循环方式重复。每次搜索后，智能体会利用记录的结果测试更好的策略，然后将改进版本应用到下一次实时运行中。在整个循环过程中，唯一变化的是搜索策略；生成解决方案的模型保持不变。","研究人员在八项任务上测试了 Dream-RSI，涵盖三个领域，使用了 Gemini 3.1 Pro：https://the-decoder.com/google-releases-gemini-3-1-pro-with-improved-reasoning-capabilities/?cmpscreencustom=1 和 Gemini 3.7 Flash：https://the-decoder.com/gemini-3-7-flash-lands-with-coding-gains-and-undercuts-its-three-week-old-predecessors-price-by-50/。每次对比使用了具有相同起始条件但固定搜索策略的基线。","其中一项任务要求系统为基因组学和金融中常用的统计计算编写运行最快的程序。Dream-RSI 的程序在所有六个测试数据集上的运行速度都比已建立的库 sklearn 和 glmnet 更快。","使用 Gemini 3.1 Pro 时，平均运行时间从 3,587 毫秒降至 2,931 毫秒，而尝试次数从 550 次降至 317 次。Dream-RSI 还超过了一种名为 SimpleTES 的竞争系统，后者需要 51,200 次运行，而 Dream-RSI 仅需 317 次尝试。","数学优化任务和高效 GPU 内核编写的努力也呈现出同样的模式，以更低的计算成本获得了相当或更好的结果。在两个 GPU 任务中，Dream-RSI 的性能相当，同时运行次数减少了最多 2.43 倍。在另外两个任务中，它在相同预算下提供了最多 2.09 倍的性能。","在后续分析中，研究人员测试了另一种使用搜索历史的方法。他们没有用它们来回放以测试策略，而是将这些历史压缩成指令，告诉代理在哪里搜索。","在一个 GPU 任务中，包含这些指令的版本表现不如没有指令的版本。研究人员建议，过于具体的指令可能会过度缩小搜索空间，阻止代理探索更广泛的方法。","同样的分析显示了学习策略如何调整其努力。当性能提高时，它最初减少了尝试次数。当进展停滞时，它再次增加搜索努力，这与进一步的收益一致。研究人员已在 GitHub 分享了代码和更多细节：https://github.com/zhengkid/Dream-RSI。","递归自我改进最近吸引了越来越多的关注。该领域的发展是 Anthropic 首席执行官 Dario Amodei 最近对人工智能研究速度发出警告的原因之一：https://the-decoder.com/ex-deepmind-vp-vinyals-says-ai-self-improvement-is-coming-but-wont-trigger-an-intelligence-explosion/。","谷歌 DeepMind 在 2025 年推出了 AlphaEvolve：https://the-decoder.com/alphaevolve-is-google-deepminds-new-ai-system-that-autonomously-creates-better-algorithms/，使用相同的基本原理。Gemini Flash 生成代码提案，Gemini Pro 分析这些提案，进化算法选择最佳版本。Dream-RSI 在这个过程之上再提升一级，通过优化搜索策略本身实现改进。","AutoTTS：https://the-decoder.com/researchers-let-claude-code-discover-ai-scaling-algorithms-that-humans-probably-wouldnt-have-designed/?cmpscreencustom=1 使用类似的方法，利用编码代理在模拟环境中搜索算法。这些算法决定语言模型何时启动、扩展或放弃推理路径。最终得到的方法在计算资源更少的情况下超越了手动设计的方法。","谷歌研究最近提出了一种用 WikiSkill 重用过去运行的方法：https://the-decoder.com/google-gives-ai-agents-their-own-wiki-so-they-can-learn-from-mistakes-and-successes/。该系统在一个 wiki 中记录失败和成功，并将其转化为可供代理使用的可重用指令。Dream-RSI 的后续分析表明，像这样的明确指令可能会限制在开放式搜索任务中的探索。","Meta 通过 Hyperagents 更进一步：https://the-decoder.com/metas-hyperagents-improve-at-tasks-and-improve-at-improving/?cmpscreencustom=1，允许代理重新编写控制它们如何提高的机制。","保持 AI 动态更新。清晰、有用、无废话。","关注 The Decoder 以获取 AI 新闻、背景故事和专家分析。","解码器：https://the-decoder.com/"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Google 与 DeepMind 研究人员提出 Dream-RSI，通过回放已完成搜索的记录，离线测试替代搜索策略；该方法只调整智能体的搜索策略，不改动底层 AI 模型。","background":"复杂任务的搜索空间可能很大，智能体需要决定继续、并行或放弃哪些路径。固定策略无法从经验中学习，实时调整策略则需要反复运行并承担较高计算成本。","viewpoint":"Aioga 判断：Dream-RSI 的重点是复用既有搜索历史来评估不同决策顺序，试图降低重复计算需求；其改进对象是探索过程，而非底层模型本身。","implications":"可能影响：该方法可能提升既有搜索记录的利用效率，但回放仅能测试已记录搜索树中的不同决策，不代表能够在回放阶段产生全新的解决方案，其适用范围仍需要结合后续研究判断。","nextStep":"后续观察：需要关注 Dream-RSI 在不同复杂搜索任务中的评估结果，以及离线策略测试能否稳定转化为后续搜索改进；现有材料不足以判断其普遍适用性。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-09-19T12:08:25.578Z","sourceHash":"532f9479f85e0779","review":{"approved":true,"groundedness":94,"clarity":91,"duplicationRisk":18,"blockingIssues":[],"notes":["“降低重复计算需求”是基于来源中“无需重复昂贵计算”的合理概括。","“Aioga 判断”已明确标注为观点，未冒充来源事实。","“可能影响”“仍需要结合后续研究判断”等表述较为审慎，与材料对方法适用范围的限制相符。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","editorial-labels","inference-boundary","low-source-overlap","no-html","independent-ai-review"]}},"tags":["行业动态","The Decoder：AI News（RSS）"],"translations":{"zh-CN":{"title":"Google 与 DeepMind 提出 Dream-RSI，让 AI 智能体通过回放搜索历史改进策略","summary":"Google 与 DeepMind 研究人员提出 Dream-RSI 方法，通过回放已完成的搜索记录离线测试数千种替代策略，只优化搜索策略而不改动底层模型。","category":"行业动态","source":"the-decoder.com","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google 与 DeepMind 提出 Dream-RSI，让 AI 智能体通过回放搜索历史改进策略 - Aioga AI资讯","description":"Google 与 DeepMind 研究人员提出 Dream-RSI 方法，通过回放已完成的搜索记录离线测试数千种替代策略，只优化搜索策略而不改动底层模型。","url":"https://www.aioga.com/news/cmu8arj6g1b6erogrovd0pigf/","articleBody":["谷歌和DeepMind的研究人员开发了一种方法，帮助AI代理更高效地处理复杂的搜索任务。该方法利用过去的搜索运行来测试新策略，而无需重复昂贵的计算。","自我改进的AI代理有望有一天能够自行发现新的算法、数学问题的解决方案或更快速的代码。它们遵循相同的基本过程：提出解决方案、评估结果、从中学习并再次尝试。经过成千上万次尝试，它们逐渐接近良好的结果。","对于复杂任务，搜索空间可能会变得极大。代理必须不断决定哪些有前景的方法值得追求，哪些可以并行尝试，以及哪些应被放弃。这个过程称为探索，它可能决定搜索是否成功，或者是否浪费计算资源去追逐错误的想法。","来自谷歌和DeepMind的一个研究团队推出了“Dream-RSI：https://github.com/zhengkid/Dream-RSI”以改进这些决策。该方法改变了智能体的搜索方式，而不是底层的人工智能模型。","现有方法通常以两种方式处理探索。固定搜索策略无法从经验中学习，因此代理可能会多次碰到相同的死胡同。在搜索过程中调整策略可以避免这种僵化，但代价不小。需要多次尝试才能确定策略是否有效，而测试无数备选方案则意味着要重复耗时且昂贵的运行。","研究人员建议重用已完成搜索的数据，在代理已经探索过的空间中测试替代策略。代理在搜索过程中记录其尝试及其结果，从而提供后续回放这些决策所需的数据。","研究人员将此方法比作在陌生区域寻找路线。第一次访问时，你会遇到死路、回头并努力寻找路径。然而，一旦你有了心理地图，就可以规划不同的路线，而无需再次访问每个地点。","Dream-RSI 将这一原理应用于记录的搜索历史。与其在实时运行中测试新策略，不如让智能体在存储的结果中运行它。这使它能够检查，如果先采用其他方法或放弃某些先前的步骤，会发生什么情况。系统不会在重放过程中完全创造新的解决方案；它是在记录的搜索树中测试不同的决策。","由于这些结果已经存在，智能体不需要再次生成或评估解决方案，从而避免了实时运行所需的昂贵计算。这使得测试新的搜索策略成本更低。研究人员称这一过程为“做梦”。智能体会演练成千上万种变体，并选择最佳方案，然后再将其应用于实时搜索。","这一过程以循环方式重复。每次搜索后，智能体会利用记录的结果测试更好的策略，然后将改进版本应用到下一次实时运行中。在整个循环过程中，唯一变化的是搜索策略；生成解决方案的模型保持不变。","研究人员在八项任务上测试了 Dream-RSI，涵盖三个领域，使用了 Gemini 3.1 Pro：https://the-decoder.com/google-releases-gemini-3-1-pro-with-improved-reasoning-capabilities/?cmpscreencustom=1 和 Gemini 3.7 Flash：https://the-decoder.com/gemini-3-7-flash-lands-with-coding-gains-and-undercuts-its-three-week-old-predecessors-price-by-50/。每次对比使用了具有相同起始条件但固定搜索策略的基线。","其中一项任务要求系统为基因组学和金融中常用的统计计算编写运行最快的程序。Dream-RSI 的程序在所有六个测试数据集上的运行速度都比已建立的库 sklearn 和 glmnet 更快。","使用 Gemini 3.1 Pro 时，平均运行时间从 3,587 毫秒降至 2,931 毫秒，而尝试次数从 550 次降至 317 次。Dream-RSI 还超过了一种名为 SimpleTES 的竞争系统，后者需要 51,200 次运行，而 Dream-RSI 仅需 317 次尝试。","数学优化任务和高效 GPU 内核编写的努力也呈现出同样的模式，以更低的计算成本获得了相当或更好的结果。在两个 GPU 任务中，Dream-RSI 的性能相当，同时运行次数减少了最多 2.43 倍。在另外两个任务中，它在相同预算下提供了最多 2.09 倍的性能。","在后续分析中，研究人员测试了另一种使用搜索历史的方法。他们没有用它们来回放以测试策略，而是将这些历史压缩成指令，告诉代理在哪里搜索。","在一个 GPU 任务中，包含这些指令的版本表现不如没有指令的版本。研究人员建议，过于具体的指令可能会过度缩小搜索空间，阻止代理探索更广泛的方法。","同样的分析显示了学习策略如何调整其努力。当性能提高时，它最初减少了尝试次数。当进展停滞时，它再次增加搜索努力，这与进一步的收益一致。研究人员已在 GitHub 分享了代码和更多细节：https://github.com/zhengkid/Dream-RSI。","递归自我改进最近吸引了越来越多的关注。该领域的发展是 Anthropic 首席执行官 Dario Amodei 最近对人工智能研究速度发出警告的原因之一：https://the-decoder.com/ex-deepmind-vp-vinyals-says-ai-self-improvement-is-coming-but-wont-trigger-an-intelligence-explosion/。","谷歌 DeepMind 在 2025 年推出了 AlphaEvolve：https://the-decoder.com/alphaevolve-is-google-deepminds-new-ai-system-that-autonomously-creates-better-algorithms/，使用相同的基本原理。Gemini Flash 生成代码提案，Gemini Pro 分析这些提案，进化算法选择最佳版本。Dream-RSI 在这个过程之上再提升一级，通过优化搜索策略本身实现改进。","AutoTTS：https://the-decoder.com/researchers-let-claude-code-discover-ai-scaling-algorithms-that-humans-probably-wouldnt-have-designed/?cmpscreencustom=1 使用类似的方法，利用编码代理在模拟环境中搜索算法。这些算法决定语言模型何时启动、扩展或放弃推理路径。最终得到的方法在计算资源更少的情况下超越了手动设计的方法。","谷歌研究最近提出了一种用 WikiSkill 重用过去运行的方法：https://the-decoder.com/google-gives-ai-agents-their-own-wiki-so-they-can-learn-from-mistakes-and-successes/。该系统在一个 wiki 中记录失败和成功，并将其转化为可供代理使用的可重用指令。Dream-RSI 的后续分析表明，像这样的明确指令可能会限制在开放式搜索任务中的探索。","Meta 通过 Hyperagents 更进一步：https://the-decoder.com/metas-hyperagents-improve-at-tasks-and-improve-at-improving/?cmpscreencustom=1，允许代理重新编写控制它们如何提高的机制。","保持 AI 动态更新。清晰、有用、无废话。","关注 The Decoder 以获取 AI 新闻、背景故事和专家分析。","解码器：https://the-decoder.com/"]},"en":{"title":"Google and DeepMind have proposed Dream-RSI, enabling AI agents to improve strategies by reviewing search history","summary":"Google and DeepMind researchers proposed the Dream-RSI method, which offline tested thousands of alternative strategies by reviewing completed search records, optimizing only the search strategy without altering the underlying model.","category":"Industry","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google and DeepMind have proposed Dream-RSI, enabling AI agents to improve strategies by reviewing search history - Aioga AI News","description":"Google and DeepMind researchers proposed the Dream-RSI method, which offline tested thousands of alternative strategies by reviewing completed search records, optimizing only the s...","url":"https://www.aioga.com/en/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:04:55.070Z"},"ja":{"title":"GoogleとDeepMindはDream-RSIを提案しており、AIエージェントが検索履歴をレビューすることで戦略を改善できるようにします","summary":"GoogleとDeepMindの研究者たちはDream-RSI法を提案しました。これは、完成した検索記録をレビューし、基礎モデルを変えずに検索戦略のみを最適化し、数千の代替戦略をオフラインでテストする手法です。","category":"業界動向","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"GoogleとDeepMindはDream-RSIを提案しており、AIエージェントが検索履歴をレビューすることで戦略を改善できるようにします - Aioga AIニュース","description":"GoogleとDeepMindの研究者たちはDream-RSI法を提案しました。これは、完成した検索記録をレビューし、基礎モデルを変えずに検索戦略のみを最適化し、数千の代替戦略をオフラインでテストする手法です。","url":"https://www.aioga.com/ja/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:04:55.466Z"},"ko":{"title":"구글과 딥마인드는 Dream-RSI를 제안했는데, 이는 AI 에이전트가 검색 기록을 검토하여 전략을 개선할 수 있도록 합니다","summary":"구글과 딥마인드 연구진은 Dream-RSI 방법을 제안했는데, 이는 완성된 검색 기록을 검토하여 수천 가지 대안 전략을 오프라인 테스트하고, 기본 모델을 변경하지 않고 검색 전략만 최적화하는 방식입니다.","category":"업계 동향","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"구글과 딥마인드는 Dream-RSI를 제안했는데, 이는 AI 에이전트가 검색 기록을 검토하여 전략을 개선할 수 있도록 합니다 - Aioga AI 뉴스","description":"구글과 딥마인드 연구진은 Dream-RSI 방법을 제안했는데, 이는 완성된 검색 기록을 검토하여 수천 가지 대안 전략을 오프라인 테스트하고, 기본 모델을 변경하지 않고 검색 전략만 최적화하는 방식입니다.","url":"https://www.aioga.com/ko/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:04.415Z"},"es":{"title":"Google y DeepMind han propuesto Dream-RSI, que permite a los agentes de IA mejorar estrategias revisando el historial de búsquedas","summary":"Investigadores de Google y DeepMind propusieron el método Dream-RSI, que probó offline miles de estrategias alternativas revisando registros de búsqueda completados, optimizando únicamente la estrategia de búsqueda sin alterar el modelo subyacente.","category":"Industria","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google y DeepMind han propuesto Dream-RSI, que permite a los agentes de IA mejorar estrategias revisando el historial de búsquedas - Aioga Noticias de IA","description":"Investigadores de Google y DeepMind propusieron el método Dream-RSI, que probó offline miles de estrategias alternativas revisando registros de búsqueda completados, optimizando ún...","url":"https://www.aioga.com/es/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:04.441Z"},"fr":{"title":"Google et DeepMind ont proposé Dream-RSI, permettant aux agents IA d’améliorer leurs stratégies en examinant l’historique de recherche","summary":"Des chercheurs de Google et DeepMind ont proposé la méthode Dream-RSI, qui testait hors ligne des milliers de stratégies alternatives en examinant les enregistrements de recherche complets, en optimisant uniquement la stratégie de recherche sans modifier le modèle sous-jacent.","category":"Industrie","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google et DeepMind ont proposé Dream-RSI, permettant aux agents IA d’améliorer leurs stratégies en examinant l’historique de recherche - Aioga Actualités IA","description":"Des chercheurs de Google et DeepMind ont proposé la méthode Dream-RSI, qui testait hors ligne des milliers de stratégies alternatives en examinant les enregistrements de recherche...","url":"https://www.aioga.com/fr/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:13.554Z"},"de":{"title":"Google und DeepMind haben Dream-RSI vorgeschlagen, das es KI-Agenten ermöglicht, Strategien durch Überprüfung der Suchhistorie zu verbessern","summary":"Google- und DeepMind-Forscher schlugen die Dream-RSI-Methode vor, die Tausende alternativer Strategien offline testete, indem abgeschlossene Sucheinträge überprüft wurden und nur die Suchstrategie optimiert wurden, ohne das zugrundeliegende Modell zu verändern.","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google und DeepMind haben Dream-RSI vorgeschlagen, das es KI-Agenten ermöglicht, Strategien durch Überprüfung der Suchhistorie zu verbessern - Aioga KI-News","description":"Google- und DeepMind-Forscher schlugen die Dream-RSI-Methode vor, die Tausende alternativer Strategien offline testete, indem abgeschlossene Sucheinträge überprüft wurden und nur d...","url":"https://www.aioga.com/de/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:13.627Z"},"pt-BR":{"title":"Google e DeepMind propuseram o Dream-RSI, permitindo que agentes de IA melhorem estratégias ao revisar o histórico de buscas","summary":"Pesquisadores do Google e da DeepMind propuseram o método Dream-RSI, que testou offline milhares de estratégias alternativas ao revisar registros de busca completos, otimizando apenas a estratégia de busca sem alterar o modelo subjacente.","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google e DeepMind propuseram o Dream-RSI, permitindo que agentes de IA melhorem estratégias ao revisar o histórico de buscas - Aioga Notícias de IA","description":"Pesquisadores do Google e da DeepMind propuseram o método Dream-RSI, que testou offline milhares de estratégias alternativas ao revisar registros de busca completos, otimizando ape...","url":"https://www.aioga.com/pt-BR/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:22.645Z"},"ru":{"title":"Google и DeepMind предложили Dream-RSI, позволяющий агентам ИИ совершенствовать стратегии, анализируя историю поиска","summary":"Исследователи Google и DeepMind предложили метод Dream-RSI, который в офлайн-режиме тестировал тысячи альтернативных стратегий путём анализа завершённых поисковых записей, оптимизируя только стратегию поиска без изменения базовой модели.","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google и DeepMind предложили Dream-RSI, позволяющий агентам ИИ совершенствовать стратегии, анализируя историю поиска - Aioga Новости ИИ","description":"Исследователи Google и DeepMind предложили метод Dream-RSI, который в офлайн-режиме тестировал тысячи альтернативных стратегий путём анализа завершённых поисковых записей, оптимизи...","url":"https://www.aioga.com/ru/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:22.829Z"},"ar":{"title":"اقترحت جوجل وديب مايند نظام Dream-RSI، الذي يمكن وكلاء الذكاء الاصطناعي من تحسين الاستراتيجيات من خلال مراجعة سجل البحث","summary":"اقترح باحثو جوجل وديب مايند طريقة دريم-RSI، التي اختبرت آلاف الاستراتيجيات البديلة دون اتصال من خلال مراجعة سجلات البحث المكتملة، مع تحسين استراتيجية البحث فقط دون تغيير النموذج الأساسي.","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"اقترحت جوجل وديب مايند نظام Dream-RSI، الذي يمكن وكلاء الذكاء الاصطناعي من تحسين الاستراتيجيات من خلال مراجعة سجل البحث - Aioga أخبار الذكاء الاصطناعي","description":"اقترح باحثو جوجل وديب مايند طريقة دريم-RSI، التي اختبرت آلاف الاستراتيجيات البديلة دون اتصال من خلال مراجعة سجلات البحث المكتملة، مع تحسين استراتيجية البحث فقط دون تغيير النموذج ال...","url":"https://www.aioga.com/ar/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:32.020Z"},"hi":{"title":"Google और DeepMind ने ड्रीम-आरएसआई का प्रस्ताव दिया है, जिससे AI एजेंट खोज इतिहास की समीक्षा करके रणनीतियों में सुधार कर सकें","summary":"Google और डीपमाइंड शोधकर्ताओं ने ड्रीम-आरएसआई पद्धति का प्रस्ताव रखा, जिसने पूर्ण खोज रिकॉर्ड की समीक्षा करके हजारों वैकल्पिक रणनीतियों का ऑफ़लाइन परीक्षण किया, अंतर्निहित मॉडल को बदले बिना केवल खोज रणनीति का अनुकूलन किया।","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google और DeepMind ने ड्रीम-आरएसआई का प्रस्ताव दिया है, जिससे AI एजेंट खोज इतिहास की समीक्षा करके रणनीतियों में सुधार कर सकें - Aioga AI समाचार","description":"Google और डीपमाइंड शोधकर्ताओं ने ड्रीम-आरएसआई पद्धति का प्रस्ताव रखा, जिसने पूर्ण खोज रिकॉर्ड की समीक्षा करके हजारों वैकल्पिक रणनीतियों का ऑफ़लाइन परीक्षण किया, अंतर्निहित मॉडल को...","url":"https://www.aioga.com/hi/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:32.116Z"},"it":{"title":"Google e DeepMind hanno proposto Dream-RSI, che consente agli agenti IA di migliorare le strategie esaminando la cronologia delle ricerche","summary":"I ricercatori di Google e DeepMind hanno proposto il metodo Dream-RSI, che offline ha testato migliaia di strategie alternative rivedendo i record di ricerca completati, ottimizzando solo la strategia di ricerca senza alterare il modello sottostante.","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google e DeepMind hanno proposto Dream-RSI, che consente agli agenti IA di migliorare le strategie esaminando la cronologia delle ricerche - Aioga Notizie IA","description":"I ricercatori di Google e DeepMind hanno proposto il metodo Dream-RSI, che offline ha testato migliaia di strategie alternative rivedendo i record di ricerca completati, ottimizzan...","url":"https://www.aioga.com/it/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:41.137Z"},"nl":{"title":"Google en DeepMind hebben Dream-RSI voorgesteld, waarmee AI-agenten strategieën kunnen verbeteren door de zoekgeschiedenis te bekijken","summary":"Google- en DeepMind-onderzoekers stelden de Dream-RSI-methode voor, die offline duizenden alternatieve strategieën testte door voltooide zoekrecords te bekijken, waarbij alleen de zoekstrategie werd geoptimaliseerd zonder het onderliggende model te veranderen.","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google en DeepMind hebben Dream-RSI voorgesteld, waarmee AI-agenten strategieën kunnen verbeteren door de zoekgeschiedenis te bekijken - Aioga AI-nieuws","description":"Google- en DeepMind-onderzoekers stelden de Dream-RSI-methode voor, die offline duizenden alternatieve strategieën testte door voltooide zoekrecords te bekijken, waarbij alleen de...","url":"https://www.aioga.com/nl/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:41.209Z"},"tr":{"title":"Google ve DeepMind, yapay zeka ajanlarının arama geçmişini inceleyerek stratejilerini geliştirmelerini sağlayan Dream-RSI'yi önermiştir","summary":"Google ve DeepMind araştırmacıları, tamamlanmış arama kayıtlarını inceleyerek binlerce alternatif stratejiyi çevrimdışı test eden ve yalnızca arama stratejisini optimize eden ve temel modeli değiştirmeden Dream-RSI yöntemini önerdi.","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google ve DeepMind, yapay zeka ajanlarının arama geçmişini inceleyerek stratejilerini geliştirmelerini sağlayan Dream-RSI'yi önermiştir - Aioga AI Haberleri","description":"Google ve DeepMind araştırmacıları, tamamlanmış arama kayıtlarını inceleyerek binlerce alternatif stratejiyi çevrimdışı test eden ve yalnızca arama stratejisini optimize eden ve te...","url":"https://www.aioga.com/tr/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:50.262Z"},"vi":{"title":"Google và DeepMind đã đề xuất Dream-RSI, giúp các tác nhân AI cải thiện chiến lược bằng cách xem lại lịch sử tìm kiếm","summary":"Các nhà nghiên cứu của Google và DeepMind đã đề xuất phương pháp Dream-RSI, thử nghiệm trực tuyến hàng nghìn chiến lược thay thế bằng cách xem xét các bản ghi tìm kiếm đã hoàn thành, chỉ tối ưu hóa chiến lược tìm kiếm mà không thay đổi mô hình cơ bản.","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google và DeepMind đã đề xuất Dream-RSI, giúp các tác nhân AI cải thiện chiến lược bằng cách xem lại lịch sử tìm kiếm - Tin tức AI Aioga","description":"Các nhà nghiên cứu của Google và DeepMind đã đề xuất phương pháp Dream-RSI, thử nghiệm trực tuyến hàng nghìn chiến lược thay thế bằng cách xem xét các bản ghi tìm kiếm đã hoàn thàn...","url":"https://www.aioga.com/vi/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:50.556Z"},"id":{"title":"Google dan DeepMind telah mengusulkan Dream-RSI, memungkinkan agen AI meningkatkan strategi dengan meninjau riwayat pencarian","summary":"Peneliti Google dan DeepMind mengusulkan metode Dream-RSI, yang secara offline menguji ribuan strategi alternatif dengan meninjau catatan pencarian yang telah selesai, mengoptimalkan hanya strategi pencarian tanpa mengubah model dasarnya.","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google dan DeepMind telah mengusulkan Dream-RSI, memungkinkan agen AI meningkatkan strategi dengan meninjau riwayat pencarian - Berita AI Aioga","description":"Peneliti Google dan DeepMind mengusulkan metode Dream-RSI, yang secara offline menguji ribuan strategi alternatif dengan meninjau catatan pencarian yang telah selesai, mengoptimalk...","url":"https://www.aioga.com/id/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:59.662Z"},"th":{"title":"Google และ DeepMind ได้เสนอ Dream-RSI ซึ่งช่วยให้เอเจนต์ AI ปรับปรุงกลยุทธ์โดยการตรวจสอบประวัติการค้นหา","summary":"นักวิจัยจาก Google และ DeepMind ได้เสนอวิธี Dream-RSI ซึ่งทดสอบกลยุทธ์ทางเลือกหลายพันแบบแบบออฟไลน์โดยการตรวจสอบบันทึกการค้นหาที่เสร็จสมบูรณ์ โดยปรับแต่งเฉพาะกลยุทธ์การค้นหาโดยไม่เปลี่ยนแปลงโมเดลพื้นฐาน","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google และ DeepMind ได้เสนอ Dream-RSI ซึ่งช่วยให้เอเจนต์ AI ปรับปรุงกลยุทธ์โดยการตรวจสอบประวัติการค้นหา - ข่าว AI Aioga","description":"นักวิจัยจาก Google และ DeepMind ได้เสนอวิธี Dream-RSI ซึ่งทดสอบกลยุทธ์ทางเลือกหลายพันแบบแบบออฟไลน์โดยการตรวจสอบบันทึกการค้นหาที่เสร็จสมบูรณ์ โดยปรับแต่งเฉพาะกลยุทธ์การค้นหาโดยไม่เป...","url":"https://www.aioga.com/th/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:05:59.705Z"},"pl":{"title":"Google i DeepMind zaproponowały Dream-RSI, umożliwiając agentom AI ulepszanie strategii poprzez przeglądanie historii wyszukiwań","summary":"Badacze Google i DeepMind zaproponowali metodę Dream-RSI, która offline testowała tysiące alternatywnych strategii, przeglądając ukończone rekordy wyszukiwania, optymalizując jedynie strategię wyszukiwania bez zmiany modelu podstawowego.","category":"行业动态","source":"The Decoder：AI News（RSS）","aggregationSource":"The Decoder：AI News（RSS）","pageTitle":"Google i DeepMind zaproponowały Dream-RSI, umożliwiając agentom AI ulepszanie strategii poprzez przeglądanie historii wyszukiwań - Aioga Wiadomości AI","description":"Badacze Google i DeepMind zaproponowali metodę Dream-RSI, która offline testowała tysiące alternatywnych strategii, przeglądając ukończone rekordy wyszukiwania, optymalizując jedyn...","url":"https://www.aioga.com/pl/news/cmu8arj6g1b6erogrovd0pigf/","contentTranslated":true,"sourceHash":"b28516ecebcd9f07","translatedAt":"2026-09-19T12:06:08.887Z"}},"evidenceTier":"verified-news","reviewStatus":"automated-ingest","indexable":true,"editorialCover":""}}