TL;DR: RARO teaches language models to reason from expert demonstrations - no task-specific verifier needed. Many real tasks don’t have an automatic "checker." RARO (Relativistic Adversarial Reasoning Optimization) uses inverse reinforcement learning to learn from examples instead. How it works: a policy tries to produce answers
Training powerful AI models needs lots of data—but real data can be scarce, costly, or sensitive. Meet Matrix, a peer-to-peer framework that makes generating high-quality synthetic data faster and easier. Instead of a central "traffic cop," Matrix lets lightweight agents talk directly by passing messages through distributed
Duplicate and lookalike items clutter online feeds and ads, hurting user experience. FITRep is an attention-guided, white-box way to represent items for fine-grained deduplication. Inspired by Feature Integration Theory, it teaches Multimodal LLMs to separate what’s primary from what’s auxiliary—so structures don’t collapse into one vague
AI
Extreme weather is getting worse, but understanding the physical “why” behind each event still takes weeks of expert detective work. Forecasting AI has improved, yet diagnostic reasoning — explaining causes — has lagged. Enter EWE (Extreme Weather Expert), a new AI agent that acts like a meteorologist-in-the-loop. EWE plans analyses, reasons in
Smarter AI Teams, Smaller Bills: Meet BAMAS Large language model (LLM) agent teams can tackle tough problems—but their cloud bills add up fast. BAMAS is a new method for building multi-agent systems that keeps performance high while respecting a budget. How it works: * Pick the right mix of models.
Matrix: Faster, flexible synthetic data—without a central bottleneck Training AI often needs synthetic data, especially when real data is scarce, pricey, or private. But most generators rely on a central “traffic cop,” slowing things down. Matrix flips the script with a peer‑to‑peer design. Tiny specialized agents pass
AI
Extreme weather is hitting harder and more often, but diagnosing the “why” behind each event is slow and expert-heavy. Meet EWE (Extreme Weather Expert): an AI agent that acts like a weather detective. * Thinks like experts: Plans analyses, reasons step-by-step, and uses a meteorology toolkit. * Works end-to-end: Turns raw atmospheric
cybersecurity
Can AI spot phishing? This new email dataset puts it to the test Phishing and spam are getting smarter—often written by large language models. Rebeka Toth, Tamas Bisztray, and Richard Dubniczky release a labeled email dataset that separates phishing, spam, and legitimate messages, and flags whether they were written
Policymakers don’t need crystal balls—they need responsible foresight. This paper introduces "responsible computational foresight": human-centric AI plus simulations to anticipate risks and opportunities, and to design policies that are sustainable, resilient, and fair. * Human-centered: AI augments judgment, not replaces it. * Systems-aware: Accounts for social, environmental, economic,
AI
AI on our phones still makes awkward mistakes—and fixing them usually means heavy retraining in the cloud. This research shows a lightweight way to correct errors on-device, in seconds. What’s new: * One-shot fixes: Show the app a single example, and it updates its prototype for that class—no
cybersecurity
Smarter email defenses, grounded in real messages Phishing and spam are evolving fast—often with help from AI. This study releases a large, carefully labeled email dataset spanning phishing, spam, and legitimate messages, with a key twist: it marks whether each message was written by a human or an LLM.
Robotics
Robots guided by Vision-Language-Action (VLA) models are getting good at everyday tasks — but most still grab with simple two-finger claws. That limits them on flat, slippery, or handle-less surfaces. VacuumVLA adds a low-cost twist: a single end-effector that combines a standard gripper with a vacuum suction tool. It can switch