Video Pretraining Cuts Surgical Robot Training Data Needs by Half
Researchers show that learning from unlabeled endoscopic footage dramatically improves surgical manipulation with limited labeled demonstrations.
Category
AI Research · Page 2
Papers, breakthroughs, benchmarks, and the long-arc trends shaping artificial intelligence. arXiv highlights and lab announcements, distilled.
Pre-Pandemic Language Patterns Predicted COVID Political Divides
New research shows partisan splits on COVID-19 were largely baked into American discourse years before the virus emerged.
The Transformer Era Faces Its Reckoning as Startups Race for LLM's Future
Nine years after their debut, transformers are showing their limits. A new wave of startups is betting they can build the next generation of language models.
Hugging Face Researchers Call for Limits on Autonomous AI Systems
A new paper argues that giving AI systems more independent control amplifies risks, arriving as the industry races to deploy autonomous agents.
AI Safety Rules Create Unequal Playing Field for Cyber Defense
Guardrails designed to prevent misuse are blocking defenders while failing to stop determined attackers, new research warns.
New Framework Uses Resource Limits to Govern Deployed AI Agents
Researchers propose a mechanism-design model treating computational budgets as the primary lever for controlling AI systems in real-world deployment.
Semantic Search With Embeddings: Architecture and Evaluation
How to build vector search systems that outperform keyword matching, and where they fail.
New Training Method Fixes Critical Imbalance in Multimodal AI Models
Researchers show how balancing text and vision processing during training significantly improves reasoning in large multimodal models while cutting computational costs.
Researchers Develop Safety Layer for Dual-Arm Robot Coordination
New intervention framework uses predictive models to improve robotic task completion while preventing unsafe actions in complex two-handed operations.
New Dataset Captures Full Human Motion to Train Embodied AI Systems
Researchers release 150 hours of synchronized video, touch, and motion data designed to close the perception-action gap in robotics training.
Fine-Tune LLMs with LoRA: When It's Worth It and How to Do It Safely
A practical guide to LoRA fine-tuning, QLoRA quantization, cost considerations, and failure modes for ML engineers.
Humanoid Robot Learns to Dodge in Real Time Using Camera Vision
New framework merges safety constraints with realistic onboard sensors to help robots evade moving objects without perfect information.
Researchers Cut AI Model Training Time in Half With Strategic Teacher Handoffs
New distillation technique lets smaller language models learn more efficiently by having teachers intervene at critical moments.
New Framework Maps Data Strategy for Training Better Robot Systems
Researchers outline how to combine real-world, simulation, and vision-language datasets to build embodied AI agents that can manipulate objects effectively.
How AI Agents Learn to Plan Across Multiple Steps
Researchers reveal the mechanics behind teaching foundation models to reason through complex, multi-stage tasks.
Researchers Identify Critical Flaw in Diffusion Model Training Method
New research reveals how a fundamental mismatch in guidance mechanisms undermines knowledge transfer in modern image generation systems.
Enterprise AI Agents Need Systems Design, Not Just Better Models
New research reveals why deploying autonomous AI workflows at scale requires rethinking infrastructure, monitoring, and capacity planning entirely.
Cornell Researchers Beam AI Model Updates Directly Into Robot Memory
New optical receiver technology could slash energy consumption in AI systems by transmitting data via light instead of power-hungry electrical connections.