Large Language Models (LLMs) have demonstrated notable capabilities across financial tasks, including financial report summarization, earnings call transcript analysis, and asset classification. However, their real-world effectiveness in managing complex fund investment remains inadequately assessed. A fundamental limitation of existing benchmarks for evaluating LLM-driven trading strategies is their reliance on historical back-testing, inadvertently enabling LLMs to "time travel"—leveraging future information embedded in their training corpora, thus resulting in possible information leakage and overly optimistic performance estimates. To address this issue, we introduce DeepFund, a live fund benchmark tool designed to rigorously evaluate LLM in real-time market conditions. Utilizing a multi-agent architecture, DeepFund connects directly with real-time stock market data—specifically data published after each model’s pretraining cutoff—to ensure fair and leakage-free evaluations. Empirical tests on nine flagship LLMs from leading global institutions across multiple investment dimensions—including ticker-level analysis, investment decision-making, portfolio management, and risk control—reveal significant practical challenges. Notably, even cutting-edge models such as DeepSeek-V3 and Claude-3.7-Sonnet incur net trading losses within DeepFund real-time evaluation environment, underscoring the present limitations of LLMs for active fund management. Our code is available at https://github.com/HKUSTDial/DeepFund.

Underwater sonar images tend to carry specific types of noise due to the unique imaging and collection approach. Existing image denoising algorithms are mostly designed for Gaussian white noise in optical scenes, and will degrade when migrated to underwater scenes. In addition, this class of algorithms requires numerous pairs of sonar images that meet their requirements, which is virtually impossible. In an innovative first, we have designed a dedicated combination for submarine sonar scenarios, namely Rapid Diffusion Probabilistic Model (RDPM). Specifically, the primary framework of RDPM is a self-encoder based on diffusion model , whose training relies exclusively on sonar images so as to reduce the complexity of combination. Compared to current algorithms, RDPM incorporates two additional limiters (i.e., global limiter and local limiter) to constrain the denoising process, which reduces its uncontrollability. Furthermore, these two limiters are trained in a separate manner, which not only dilutes the fine-tuning cost and model size, making it easier to deploy, but also facilitates the migration to different sea areas. Adequate experiments demonstrate that our proposed method outperforms unpaired denoising models on several publicly available sonar image datasets, and excels in controllability and migratability. Our code is available at https://github.com/zbyhnu/LimitDiffusionModel.git .