Opening the AI Black Box: A Path to Verifiable Safety
New research proposes a rigorous statistical framework for certifying the safety of artificial intelligence systems, moving beyond abstract risk assessments.
Science
Predicting Sepsis Before It Strikes: A New Approach to Early Warning
Researchers have developed a novel system that forecasts the onset of sepsis by simulating a patient’s physiological trends, offering a crucial window for intervention.
Can AI Really Analyze Stocks?
A new benchmark reveals the current capabilities-and limitations-of artificial intelligence in conducting professional financial equity research.
Tracking the Tipping Point: How Network Analysis Can Predict Market Crashes
![A network analysis of co-trading relationships reveals distinct communities of principal participants-identified through comprehensive time-series data ([latex]DNM1[/latex]) and those focused on trading point processes ([latex]DNM2[/latex])-with participants unique to each approach highlighted in blue and red, respectively, while those present in both are shown in orange, and network layouts either maintain consistency with prior visualizations or prioritize proximity between tightly connected nodes.](https://arxiv.org/html/2604.21297v1/x6.png)
New research reveals how analyzing the interactions of traders can provide early warnings of financial instability, moving beyond traditional economic indicators.
The AI Trader: Remaking Finance with Intelligent Agents
A new generation of artificial intelligence is poised to reshape financial markets, but realizing its potential requires careful consideration of emerging risks and regulatory challenges.
Predicting the Future: A New Dataset for Forecasting Markets
Researchers have released a comprehensive dataset from the Polymarket platform, enabling deeper analysis of prediction market dynamics and potential improvements to economic forecasting.
Decoding ESG Data with AI
![A system-ESGLens-processes sustainability reports from major market indices-QQQ, S&P 500, and Russell 1000-through a five-stage pipeline of data collection, PDF processing utilizing [latex]FAISS[/latex] vector databases and [latex]OpenAI[/latex] embeddings, targeted data extraction guided by GRI standards, ChatGPT-driven summarization, and ultimately, regression-model-either Neural Network or LightGBM-based scoring to generate a quantitative ESG assessment benchmarked against existing LSEG data, demonstrating an attempt to distill complex qualitative information into a measurable, comparable metric subject to the inherent decay of any derived score.](https://arxiv.org/html/2604.19779v1/01-Fig/NLP_1-2_detailed-process.png)
A new framework uses artificial intelligence to automatically analyze corporate sustainability reports and predict ESG performance.
Stress-Testing AI: A New Framework for Security Evaluation
Researchers have developed an automated system to proactively identify weaknesses in artificial intelligence systems before malicious actors can exploit them.
Coastal Resilience: Forecasting Storm Surge with Spatial Intelligence
![The system dissects hurricane storm surge forecasting by constructing a spatio-temporal graph-nodes representing gauge stations and edges quantifying their correlations-to predict localized water level offsets [latex]\hat{o}_{i}(t)[/latex] and refine physics-based ADCIRC models, effectively learning to correct inherent biases in surge prediction.](https://arxiv.org/html/2604.20688v1/x2.png)
A new graph neural network model, StormNet, leverages connections between coastal monitoring stations to significantly reduce biases and improve the accuracy of storm surge predictions.
Predictive Heat Exchanger Maintenance: A Faster Path to Reliability
![The heat-exchanger model leverages prior probability densities-established for the changepoint time τ, fouling strength [latex]\beta_f[/latex], leak rate [latex]\beta_l[/latex], and fouling-event arrival rate λ-to constrain parameter estimation within the scenarios detailed in Table 1, acknowledging the inherent uncertainty in predicting system degradation.](https://arxiv.org/html/2604.20735v1/x3.png)
New research demonstrates a computationally efficient method for monitoring heat exchanger health, paving the way for real-time diagnostics and scalable predictive maintenance programs.