Mining Research for Open Source Survival
A new approach automatically extracts practical advice from academic studies to help open-source projects thrive.
Science
Securing Web3 Transactions with Sovereign AI
A new Edge-Local-Cloud architecture, DMind-3, promises safer, faster, and more private transaction execution by prioritizing user control and robust verification.
When AI Disagrees: Measuring Fairness Under the EU’s New Rules
The EU AI Act aims to regulate high-risk systems, but accurately assessing individual performance becomes complex when multiple equally valid AI models offer differing predictions.
Stabilizing Clinical Predictions: A Boostrap Approach
New research demonstrates a method for reducing unpredictable variations in clinical risk scores without compromising accuracy or transparency.
After the Flames: Predicting Deadly Post-Wildfire Mudflows
New research leverages machine learning and controlled experiments to forecast the onset of dangerous debris flows in burn scars, offering critical insights for mitigation efforts.
Rewiring AI for Safety: A Neuron-Level Approach
Researchers are exploring how to make large language models safer by directly intervening at the level of individual neurons.
Beyond Token Confidence: Predicting Failure in AI Agents
![The system assesses agentic reasoning through trajectory-level uncertainty estimation, quantifying risk at each step by combining content-aware surprisal, repetition detection, action-observation mismatch, and user-agent coordination gaps-calculated as [latex]r\_{t}=\max(U\_{t},\alpha D\_{a}(t),\beta D\_{o}^{A}(t),\gamma D\_{o}^{U}(t))[/latex]-and then aggregates these risks using tail-focused metrics like top-K mean and the [latex]\ell\_{\in fty}[/latex] norm to identify potential failures in multi-turn interactions.](https://arxiv.org/html/2602.11409v1/figures/overview.png)
New research introduces a method for assessing the risk of multi-step AI reasoning, moving beyond simple token-level probabilities to evaluate entire decision pathways.
Bridging the Silos: How Digital Ecosystems Can Unlock Collaboration
A new framework explores how interconnected digital networks can overcome fragmentation and foster cooperation in an increasingly divided world.
Beyond Firewalls: AI Agents for a Self-Defending Cyber Future
A new architecture proposes shifting cybersecurity from reactive pipelines to proactive, self-governing AI agents that can adapt to evolving threats.
Can Machine Learning Models Accurately Simulate Global Climate?
A new study assesses the performance of a hybrid machine learning-physics model in replicating key atmospheric phenomena and its limitations under extreme warming.