Artificial Intelligence (AI) benchmarks play a central role in measuring progress in model development and guiding deployment decisions. However, many benchmarks quickly become saturated, meaning that they can no longer differentiate between the best-performing models, diminishing their long-term value. In this study, we analyze benchmark saturation across 60 Large Language Model (LLM) benchmarks selected from technical reports by major model developers. To identify factors driving saturation, we characterize benchmarks along 14 properties spanning task design, data construction, and evaluation format. We test five hypotheses examining how each property contributes to saturation rates. Our analysis reveals that nearly half of the benchmarks exhibit saturation, with rates increasing as benchmarks age. Notably, hiding test data (i.e., public vs. private) shows no protective effect, while expert-curated benchmarks resist saturation better than crowdsourced ones. Our findings highlight which design choices extend benchmark longevity and inform strategies for more durable evaluation.
Long-horizon embodied planning is challenging because the world does not only change through an agent's actions: exogenous processes (e.g., water heating, dominoes cascading) unfold concurrently with the agent's actions. We propose a framework for abstract world models that jointly learns (i) symbolic state representations and (ii) causal processes for both endogenous actions and exogenous mechanisms. Each causal process models the time course of a stochastic cause-effect relation. We learn these world models from limited data via variational Bayesian inference combined with LLM proposals. Across five simulated tabletop robotics environments, the learned models enable fast planning that generalizes to held-out tasks with more objects and more complex goals, outperforming a range of baselines.
The complement of an arrangement of diagonal subspaces x_i_1=…=x_i_k in the real space is defined by a simplicial complex 𝒦 . In this paper, we prove that the complement of a diagonal subspace arrangement is homotopy equivalent to a subcomplex Perm(𝒦) of faces of the permutohedron. The product in the cohomology ring of the complement of a diagonal arrangement is then described via Saneblidze and Umble’s cellular approximation of the diagonal map in the permutohedron. We consider the projection from the permutohedron to the cube and prove that the Saneblidze–Umble diagonal is mapped to the diagonal constructed by Li Cai for describing the product in the cohomology of a real moment–angle complex.
This essay examines the automation development of China's Yangshan Port by benchmarking South Korea's Busan Port within the broader context of international maritime trade and port modernization. As global trade continues to expand, ports are increasingly required to improve efficiency, capacity, and technological integration in order to remain competitive.With the development of automation， many companies reduce investment in improving environmental protections and operational transparency. Therefore, in order to pursue long-term efficiency, a more comprehensive evaluating framework is needed. This study argues that port competitiveness is no longer determined solely by throughput or physical scale, but increasingly by the balance between Economic, Environmental, Social, and Governance (EESG) factors.Using a comparative EESG framework, this paper analyzes Yangshan Port and Busan Port across four dimensions: economic automation performance, environmental sustainability, social impact, and governance structure. The study highlights that Yangshan Port has a significant advantage in automation level and productive efficiency, while Busan port is more focused on environmental, social and governance factors. The findings suggest that although Yangshan Port demonstrates superior economic efficiency, Busan Port offers a more balanced EESG performance. Ultimately, we discover that there is a significant trade-off between efficiency and managing social and environmental externalities. Yangshan Port should refer to the Busan regulation and pollution solving model to strengthen environmental, social, governance(ESG) aspects.
This study develops and evaluates a dynamic downside-minimizing portfolio through five stages. Starting with a 4-asset universe of equity, bonds, gold, and a volatility instrument, the portfolio is expanded to 13 and then 21 assets while incorporating mean-semivariance optimization, Asian arithmetic-average puts, American puts valued with Longstaff–Schwartz Least-Squares Monte Carlo, covered calls, momentum tilts, and regime-aware rebalancing. The final stage targets a 15% expected return. Portfolio performance is evaluated using correlated Geometric Brownian Motion with 20,000 Monte Carlo paths over a one-year horizon. The four-asset semivariance-optimized portfolio has a Sortino ratio of 2.04 and a maximum simulated loss of −23.7%, compared with 1.20 and −45.9% for SPY. Adding Asian and American puts increases the Sortino ratio to 2.26 and reduces maximum simulated loss to −8.8%. The 13-asset buy-and-hold portfolio reaches a mean Sortino ratio of 5.81. Among the rebalancing strategies tested, calendar rebalancing produced the highest mean Sortino ratio of 9.62. The 21-asset portfolio with a 15% return target has a Sortino ratio of 4.06 and maximum simulated loss of −28.8%; the full option overlay reduces the latter to −9.1%. These results characterize the tradeoffs among diversification, downside protection, return targets, and adaptive rebalancing within the simulated framework.
