Reinforcement Learning
Ray Summit 2026 Highlights AI Advances in Reinforcement Learning
Ray Summit 2026 showcased cutting-edge AI breakthroughs, from NVIDIA's RL scaling to Lila Sciences' physical AI pipelines. Here's what mattered.
SkyRL Adopts FP8 for RL, Cuts Rollout Time by 23%
SkyRL's FP8 reinforcement learning stack matches BF16 convergence while reducing rollout step time by up to 23%, boosting efficiency on NVIDIA GPUs.
Ray 2.58 Unveils Native gVisor Sandboxing for Scalable AI Workloads
Ray 2.58 introduces native gVisor sandboxing, enabling scalable, isolated environments for reinforcement learning and agentic AI workloads.
NVIDIA Vera Rubin Optimizes AI Post-Training Efficiency
NVIDIA’s Vera Rubin platform sets a new benchmark for AI post-training, maximizing intelligence per dollar in the agentic AI era.
NVIDIA Explores RLVR for AI Agents with Nemotron 3 Super
NVIDIA showcases Nemotron 3 Super and RLVR techniques to improve AI agents' domain-specific workflows, pushing reinforcement learning's practical limits.
SkyRL Adds Vision-Language RL Support for Multimodal Models
SkyRL introduces vision-language reinforcement learning, enabling scalable training for multimodal tasks. Learn how this impacts AI development.
NVIDIA Unveils AI Agent Training Method Using Synthetic Data and GRPO
NVIDIA's new approach combines synthetic data generation with reinforcement learning to train CLI agents on a single GPU, cutting training time from months to days.
Leveraging Reinforcement Learning for Scientific AI Agents
Explore how reinforcement learning enhances scientific AI agents, reducing the burden of repetitive tasks and fostering innovation, as detailed by NVIDIA.
TorchForge RL Pipelines Now Operable on Together AI's Cloud
Together AI introduces TorchForge RL pipelines on its cloud platform, enhancing distributed training and sandboxed environments with a BlackJack training demo.
NVIDIA's ProRL v2 Advances LLM Reinforcement Learning with Extended Training
NVIDIA unveils ProRL v2, a significant leap in reinforcement learning for large language models (LLMs), enhancing performance through extended training and innovative algorithms.