Bridging the African Economics Knowledge Gap
A new dataset reveals that large language models struggle with specialized African economic data, highlighting the need for enhanced retrieval mechanisms.
Science
When AI Gets Facts Wrong: A New Stress Test for High-Stakes Systems
Researchers have created a challenging benchmark to expose how easily large language models can produce incorrect information in critical fields like healthcare, finance, and law.
The Human Filter: Can Translators Spot AI-Written Text?
A new study reveals that professional translators struggle to reliably differentiate between human-authored and machine-generated Italian, raising concerns about the potential for undetectable synthetic content.
The AI Sentiment Shift: When Accuracy Comes at a Cost
New research reveals that while AI transformers excel at understanding sentiment, they increasingly exhibit a tendency to amplify polarization and lose objectivity in their assessments.
Seeing the Unseen: Synthesizing X-ray Dark-field Images with AI
Researchers have developed a new deep learning method to generate realistic dark-field radiographs from standard X-ray images, unlocking potentially valuable diagnostic information.
Reconstructing the Sun’s Hidden Magnetic Fields
![MagNet processes inputs of Hα and line-of-sight data to generate magnetic fields, [latex]B’_x[/latex] and [latex]B’_y[/latex], demonstrating how even the most meticulously constructed models are ultimately limited by the data upon which they rely.](https://arxiv.org/html/2601.15926v1/MagNet_workflow.png)
A new machine learning model accurately fills in gaps in historical solar magnetic field data, offering a powerful tool for studying past solar activity.
When AI Gets It Wrong: Spotting falsehoods in Language Models
A new framework uses principles from the human brain to identify when large language models stray from factual grounding and generate misleading information.
The Self-Improving Algorithm: Learning on the Fly
![Through iterative refinement during evaluation, a language model honed its problem-solving capabilities-demonstrated by a progression from initial reward distributions at step 0 to a final state at step 49-eventually exceeding the performance of established human baselines, all while operating with a fixed sampling budget comparable to that of a nearest-neighbor search [latex]\pi\_{\theta\_{i}}[/latex].](https://arxiv.org/html/2601.16175v1/x1.png)
A new approach allows models to refine their problem-solving skills during testing, unlocking significant performance gains across diverse scientific fields.
Watching the Markets: An AI for Spotting Manipulation
![The distribution of composite window integrity scores exhibits a heavy tail, indicative of substantial outliers in [latex]\phi_1[/latex] during early-window baseline estimation, and suggests that rank-based analysis offers a more resilient approach to comparative assessment than absolute scoring.](https://arxiv.org/html/2601.15304v1/figures/score_histogram.png)
A new system leverages publicly available data and transparent scoring to detect potentially manipulative activity in financial markets.
Fighting Fakes: AI Learns to Spot Deepfake Forgeries
A new approach combines image blending with reinforcement learning to automatically generate explanations and improve the accuracy of deepfake detection systems.