Anthropic and OpenAI have reported substantial large language model assistance in their own research and development without claiming full recursive self-improvement, or RSI, where an AI system autonomously designs and trains its successor. Anthropic’s September 2026 update showed Claude authoring over 80% of merged code and handling 26% of core R&D tasks autonomously—up sharply from early 2026—while stating no measured workflow runs fully closed-loop. OpenAI reached its “automated research intern” milestone in September, logging 3.1 agent-workdays per human workday, but retains a March 2028 target for a fully automated AI researcher. Co-founder Jack Clark shifted his personal RSI timeline forward to 2027 based on rapid gains in research-taste benchmarks, and both labs emphasize safety limits and human oversight amid calls for frontier pacing and potential regulation. These verified capability jumps without an announced closed loop shape trader views on near-term public declarations.
如果 OpenAI 或 Anthropic,或其中任一公司的官方代表,在指定日期的 23:59 ET 之前宣布该公司已实现递归自我改进,则本市场将判定为“是”。否则,本市场将以“否”结算。 递归自我改进是指公司公告所描述的一种能力:公司的 AI 自主开展所需的研究与开发,以实质性提升其最强或接近前沿水平的 AI 系统的通用能力,或制造能力更强的后继系统;且改进后的系统能够推动进一步的此类改进。人类可以设定高层次目标,并提供资源、监督和批准,但不得提供实质性的研究选择或技术解决方案。玩具式、范围狭窄、在沙盒环境中进行、概念验证性质,或仅限于提示词、技能或基准测试分数的改进,均不符合条件。 符合条件的公告必须以官方身份说明该能力已在当前实现。声称 AI 正在协助或加速公司的研究、或为该研究编写代码的说法,关于未来计划或预期的说法,猜测,采访者的描述,泄露信息,模型输出,以及第三方说法,均不符合条件。无需公开发布、具名模型、技术演示或独立验证。 此市场判定的主要依据将是 OpenAI 和 Anthropic 或其官方代表发布的官方信息;不过,也可采用可信报道形成的共识。
机器翻译。如有争议,以原文为准。