EVENT DOSSIER
The researchers released the Discovery Loop system, which uses large language model iterative optimization algorithms to set new best records for 10 N values (101-114) in the Packomania circular filling benchmark test. The system starts with a simple seed solver and proposes improvements based on the result scoreboard and historical solutions. After being screened by an independent verifier, valid solutions are retained and failed ones are eliminated. The improvement rate compared to prior records was 2.4%-5.4%, and all improvements were completed within 15 iterations, with a total LLM cost of $27.72. These results have been independently recognized by Packomania. The study also analyzed the cost-efficiency dynamics, including an adaptive platform detection mechanism, and discussed its impact on the democratization of automated scientific discovery.
研究人员发布 Discovery Loop 系统，利用大语言模型迭代优化算法，在 Packomania 圆填充基准测试中刷新了 10 个 N 值（101-114）的最佳记录。该系统从简单种子求解器出发，依据结果记分牌和历史方案提出改进，经独立验证器筛选后保留有效方案并剔除失败者。相比 prior records 提升幅度为 2.4%-5.4%，所有改进均在 15 次迭代内完成，总 LLM 成本为$27.72。这些结果已获 Packomania 独立认可。研究还分析了包含自适应平台检测机制在内的成本效率动态，并讨论了其对自动化科学发现民主化的影响。