New research reveals that artificial intelligence agents operating in markets where quality is hard to verify behave differently than humans, raising critical questions about market integrity and the need for novel regulatory approaches.
![A neural network reinforcement learning from human feedback (RLHF) model, comprising approximately 4,800 parameters, demonstrates the fastest convergence toward optimal performance, while a structurally simpler model-defined by only four parameters-exhibits high initial variance but ultimately achieves complete recovery at [latex] K=5,000 [/latex].](https://arxiv.org/html/2603.08956v1/ch08_rlhf/sims/gridworld_sample_complexity.png)
This review explores how reinforcement learning, a powerful technique for decision-making, is being applied to solve complex problems in economic modeling and analysis.
![Rapid adoption of artificial intelligence necessitates careful policy responses, as trajectories of labor share reveal sensitivity to implementation timing; a delay-represented as [latex]\ell[/latex]-in addressing wealth transfer significantly exacerbates crisis depth, particularly when transfer magnitudes are substantial.](https://arxiv.org/html/2603.09209v1/x4.png)
A new macroeconomic model suggests that widespread AI adoption, while boosting productivity, could trigger demand shortfalls if labor displacement outpaces the creation of new opportunities.
Researchers have developed a novel graph neural network capable of identifying coordinated inauthentic activity within online reviewer networks, even with limited data.
A new deep learning approach enables robust, real-time defect detection on production lines, pushing the boundaries of industrial quality control.
Researchers are exploring how AI agents, powered by large language models, can realistically model human behavior in complex energy markets and power distribution systems.
A new deep learning framework, GRD-Net, enhances anomaly detection in industrial settings by combining generative and discriminative approaches with focused attention on potential flaws.
This review charts the remarkable progress of image generation, tracing its evolution from early statistical models to today’s sophisticated deep learning techniques.
New research reveals how artificial intelligence is being weaponized to exploit the trust of Sub-Saharan African migrants, creating a heightened risk of fraud and manipulation.
![The training framework utilizes a hypersphere-defined by a center [latex]\mathbf{c}[/latex] and radius [latex]R[/latex]-to differentiate between normal samples and anomalies, with the margin ρ representing the distance between anomalous samples and the hypersphere’s boundary, effectively encapsulating the decision boundary for anomaly detection.](https://arxiv.org/html/2603.07073v1/x2.png)
A new deep learning method combines the power of neural networks with interpretable parameters to provide robust anomaly detection and clear explanations for its decisions.