AI
Ever heard of the “lottery ticket” idea in AI? It says big neural nets hide small subnetworks that can perform just as well. This paper proves a strong version of that for the heart of Transformers: multi-head attention (MHA). * The big claim: Inside a randomly initialized MHA, there exists a
AI
Ever heard of the “lottery ticket” idea in AI? It says big neural nets hide small subnetworks that can perform just as well. This paper proves a strong version of that for the heart of Transformers: multi-head attention (MHA). * The big claim: Inside a randomly initialized MHA, there exists a
AI
Struggling to compare or improve your automations? Opus offers a simple, quantitative way to score and optimize any Workflow—so you can pick what works, fix what doesn’t, and waste fewer resources. What’s inside * Opus Workflow Reward: a probabilistic score that blends likelihood of success, resource usage, and
AI
What if human pose tech could guide doctors in scans? MedSapiens takes a simple but bold idea: adapt a powerful human pose foundation model to find anatomical landmarks—key points—in medical images. * Built by fine-tuning Sapiens across multiple medical datasets. * Sets a new state of the art in success
AI
Getting AI to reflect human preferences usually means showing it lots of examples—which is slow and costly. This paper proposes a smarter path: combine the scale of RLHF (used to tune large language models) with the efficiency of Bayesian preference optimization (PBO), which actively chooses the most informative questions
RemoteSensing
AI spots offshore platforms from space—better with synthetic data Offshore wind, oil & gas, and aquaculture are growing fast. Researchers trained a YOLOv10 model on ESA Sentinel‑1 radar images from the Caspian Sea, South China Sea, Gulf of Guinea, and Brazil—then boosted learning with synthetic examples to
AI assistants are getting better at healthcare tasks—but are they safe when it comes to medications? A new study introduces RxSafeBench, a realistic test to see whether large language models (LLMs) can spot medication risks during simulated doctor-patient chats. Using a curated RxRisk DB (6,725 contraindications, 28,781
AI
LLMs are starting to act as “agents,” making choices in dynamic settings—but they often struggle with exploration vs. exploitation and rack up high regret (missed reward vs. the best strategy in hindsight). A new approach, Iterative Regret-Minimization Fine-Tuning (Iterative RMFT), trains models to decide better by learning from their
LLM
Can an LLM judge the best playlist, not just the next song? Recommender systems often serve slates—ordered lists like your home feed or a playlist. Modeling what a person prefers across domains is hard. This study tests Large Language Models as a 'world model' of user preferences:
AI
Researchers introduce Jr. AI Scientist, an autonomous "junior researcher" that can read a baseline paper from a human mentor, spot limitations, propose improvements, run experiments, and draft a new paper. Unlike earlier "fully automated" attempts, it follows a clear research workflow and uses modern coding agents
Robots learning from human videos—without the bad habits Human demos are cheap and plentiful, but humans and robots have different bodies—so copying motions often fails. X-Diffusion turns this into a strength. The trick: add noise to actions. As noise increases, tiny execution details vanish while the high-level intent
AI
GPS apps nail the fastest route, but miss real-life context: pick up meds, avoid school zones, or grab a coffee before it closes. PAVe (Personalized Agentic Vehicular Routing) adds that missing layer. It pairs proven pathfinding with an AI agent that understands your tasks and preferences. First, it computes candidate