一项新的研究论文介绍了EdgeBench，这是一个用于研究AI代理在部署后如何从真实世界环境中学习的框架。对134个任务中38,000小时代理交互的分析揭示了性能的对数-S形标度律，R^2 = 0.998，并表明代理学习速度大约每三个月翻一番。研究人员正在发布51个任务和评估框架，以促进该领域的进一步研究。 AI
影响 提供了一个框架和数据来理解和加速AI代理在部署后的学习，可能影响未来的AI开发和能力。
排序理由 研究论文，详细介绍了关于AI学习的新框架和发现。
AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →
arXiv:2607.05155v1 Announce Type: new Abstract: Pretraining scaling laws reveal that model capability improves predictably with data and compute. But learning from real world environments after deployment remains far less understood. Analyzing roughly 38,000 hours of agent intera…
Pretraining scaling laws reveal that model capability improves predictably with data and compute. But learning from real world environments after deployment remains far less understood. Analyzing roughly 38,000 hours of agent interaction with the environment across 134 real world…
Analysis of 38,000 hours of real-world agent interactions reveals log-sigmoid scaling laws for performance and exponential learning speed improvements across 134 diverse tasks.
<table> <tr><td> <a href="https://www.reddit.com/r/singularity/comments/1ulvipo/edgebench_reveals_the_next_scaling_law_onthefly/"> <img alt="EdgeBench Reveals the Next Scaling Law: On-the-Fly AI Learning Speed Doubles Every 3 Months" src="https://preview.redd.it/wzogjmyivvah1.png…