what's new in generative ai

evaluates and improves LLMs for complex inpatient decision-making by moving beyond question-answering to real clinical workflows. practitioners need this because it addresses the gap between medical benchmarks and the messy, multi-step reasoning required in actual hospital settings.

Roger Creus Castanyer, Geoffrey Bradway, Lorenz Wolf +3

uses population-based self-play with LoRA adapters to improve reasoning in LLMs through reinforcement learning with verifiable rewards. practitioners benefit because it shows how to efficiently fine-tune models for complex reasoning tasks without full retraining.

Prior Labs Team

scales foundation models for tabular data to handle 1M+ rows while reducing training and inference time significantly. this is critical for practitioners because most real-world prediction problems use tabular data, and this removes a major bottleneck in deploying models on large datasets.

Qwen Team

presents a unified model that handles vision, language, and robotic action across different tasks and robot types, reducing fragmentation in embodied AI. this matters because practitioners building robots can now use a single foundation model instead of task-specific systems, improving generalization.

combines foundation models with symbolic reasoning to enable robust industrial control in unpredictable, data-scarce environments. practitioners in manufacturing benefit because it shows how to move beyond large labeled datasets while maintaining explainability and error detection.

demonstrates an end-to-end AI-generated control system for autonomous drones with minimal human input, using LLMs and hybrid reasoning. practitioners in robotics and autonomous systems see how generative models can reduce manual engineering effort in system design and validation.

Anuj Sadani, Deepak Kumar

proposes dynamic tool gating and lazy schema loading to reduce overhead when agents manage many tools via the model context protocol. this directly addresses scalability pain points for teams building multi-tool agentic systems where tool selection and loading become bottlenecks.