RSIGym:让AI Agent自主训练提升SWE-bench性能至50.33%
Recommended read. And I agree that the RSI is also a systems engineering problem…
Agent 工程落地的硬核参考,展示了通过专用研究环境(RSIGym)显著提升 SWE-bench 分数的具体实践,值得做 Agent 的同学关注其架构思路。
Recommended read. And I agree that the RSI is also a systems engineering problem.
推荐阅读。我也同意 RSI 也是一个系统工程问题。
Self-improving agents need better research environments.
自我改进的智能体需要更好的研究环境。
RSIGym gives a research agent training, inference, evals, and sandboxes as services it can call. The agent spends its budget on experiments instead of rebuilding infra.
RSIGym 提供研究智能体的训练、推理、评估和沙盒作为可调用服务。智能体将预算用于实验,而不是重建基础设施。
With Opus 5 as the researcher, the improved system went from 17.67% to 50.33% on SWE-bench Verified.
以 Opus 5 作为研究者,改进后的系统在 SWE-bench Verified 上的表现从 17.67% 提升到了 50.33%。
Also cool to see a way to measure the quality of co-evolution between harnesses and models, which is how full-stack AI companies stay on the frontier.
看到一种衡量工具链与模型之间协同进化质量的方法也很酷,这正是全栈 AI 公司保持前沿地位的方式。
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力