谷歌“开始进入RSI跑步机”，联合创始人谢尔盖·布林在Gemini 4 Argon圆桌会议上表示，Business Insider报道称。谷歌于周三分享了该视频。布林指出，谷歌在Gemini 4之前发布的Flash模型是这一趋势的证据。
Google is “starting to get onto that RSI treadmill”, cofounder Sergey Brin said in a roundtable on Gemini 4 Argon, Business Insider reported. Google shared the video on Wednesday. Brin pointed to the Flash models Google released before Gemini 4 as evidence.
“如今，我们的研究人员和模型协同工作，以打造模型的下一个版本。”谢尔盖·布林，谷歌联合创始人递归自我改进（RSI）指的是AI帮助构建更好的AI。这将加速新模型的发布速度。布林表示，到人们观看该视频时，谷歌内部将领先几个模型版本。
“And now our researchers and the model work hand in hand to make the next version of the model.”Sergey Brin, Google cofounder Recursive self-improvement, or RSI, means AI helps build better AI. That speeds up the pace of new models. Brin said Google would be a couple of models ahead internally by the time people watched the video.
布林与谷歌的乔希·伍德沃德、图尔西·多希和梅尔文·约翰逊一起参加了此次对话。谷歌本周公开发布了Argon。布林自回归谷歌以来一直在参与Gemini项目的工作。谷歌的安全报告内容谷歌DeepMind本月发布了Argon的安全与安全报告。报告指出：“未发现AI研发的端到端显著加速，无论是AI驱动还是其他方式。”报告将AI驱动的AI研发风险评为“目前较低”。
Brin joined Google’s Josh Woodward, Tulsee Doshi and Melvin Johnson for the conversation. Google released Argon publicly this week. Brin has been working on Gemini since his return to Google. What Google’s safety report says Google DeepMind published a safety and security report for Argon this month. It finds “no significant end-to-end acceleration of AI R&D, AI-driven or otherwise”. It rates the risk from AI-driven AI R&D as “currently low”.
报告称，AI目前正在自动化AI研究的一些部分。谷歌为此调研了12位高级领导和研究人员。
The report says AI now automates some parts of AI research. Google surveyed 12 senior leaders and researchers for it.
“AI代理目前充当高杠杆助手，负责编写大部分生产代码和修复漏洞，尽管代码更改在提交前仍需接受人工审查。”Gemini 4 Argon安全与安全报告根据报告，谷歌DeepMind目前超过75%的生产代码由AI编写。完全委托的拉取请求比例从4月的7.5%上升至9月的11.9%。约88%的代码仍需要反复的人工干预或由人类编写。
“AI agents now operate as high-leverage assistants that author a majority of production code and bug fixes, though code changes remain subject to human review before submission.”Gemini 4 Argon Safety and Security Report Most production code at Google DeepMind is now more than 75% AI-written, according to the report. The share of fully delegated pull requests rose from 7.5% in April to 11.9% in September. About 88% still need repeated human steering or code written by a human.
报告指出，人类仍然是选择研究方向、设计新实验和诊断训练问题的关键。Argon仅能解决谷歌内部AI研究任务测试中大约一半的挑战。实现全自动化需要接近完美的得分。
Humans remain key to choosing research directions, designing new experiments and diagnosing training problems, the report says. Argon solves only around half of the challenges in Google’s internal test of AI research tasks. Full automation would need a near-perfect score.
谷歌通过一个由92个基准测试构建的内部指数来追踪Gemini改进的速度。在2026年4月1日之前，该指数每月上升0.091个点。在此日期之后，每月上升0.118个点，报告称这一差异在统计上不显著。
A steady pace Google tracks the rate at which Gemini improves with an internal index built from 92 benchmarks. Before 1 April 2026, it rose by 0.091 points a month. After that date, it rose by 0.118 points a month, a difference the report calls not statistically significant.
氩气位于这一趋势之上。谷歌怀疑这是由于投入增加和训练算力增加所致，“而非反映人工智能驱动的加速”。‘新档位’维汉·贾恩（Vihan Jain）是谷歌DeepMind旗下Gemini项目中RSI和科学AI的联合负责人。他在X平台上写道，该模型“嵌入在其自身的演化中”。
Argon sits above that trend. Google suspects this is due to more investment and more training compute, “rather than reflecting AI-driven acceleration”. ‘A new gear’Vihan Jain co-leads RSI and AI for science for Gemini at Google DeepMind. On X, Jain wrote that the model is “embedded in its own evolution”.
“当你的后训练需求从索要原始芯片转变为索要氩气配额时，你就知道RSI飞轮已经进入了一个新档位。”维汉·贾恩，谷歌DeepMind格莱斯顿AI（Gladstone AI）本周采访了Anthropic、OpenAI和谷歌DeepMind的研究人员，针对一份刚刚发布的报告。大多数受访者认为RSI将在三到十二个月内启动。他们表示，假设没有美国政府干预或放缓，RSI可能会更早出现。
“When your post-training demand shifts from asking for raw chips to asking for Argon quota, you know the RSI flywheel has reached a new gear.”Vihan Jain, Google DeepMind Gladstone AI interviewed researchers at Anthropic, OpenAI and Google DeepMind for a report published this week. Most believe RSI will start within three to 12 months. It could come sooner, they said, assuming no US government intervention or slowdown.
上个月，美国众议员罗·坎纳（Ro Khanna）表示他计划提出一项法案，以禁止自我改进的人工智能。OpenAI曾呼吁美国在全球范围内为其制定标准。
Last month, US congressman Ro Khanna said he plans a bill to ban self-improving AI. OpenAI has asked the US to lead global standards for it.