#reinforcement learning
Reinforcement Learning: 35 AI articles covering reinforcement learning news, analysis, and research
Articles
Left Alone With Nothing to Do, AI Agents Started Playing — AndNEW⭐9
MIT researchers let 13 AI agents roam a simulated island with no task for 30 hours. They began inventing their own games—and learned skills that helped them out...
Turning Database Grading into RL Rewards: Post-Training AgentsNEW⭐9
Microsoft's ThinkingBox-Bench turns deterministic database grading into verifiable RL rewards, bridging evaluation and post-training for AI agents.
Navalia: Advancing Cowork Agents through Verifiable Task⭐8
Navalia combines verifiable task synthesis with long-horizon post-training to build Cowork Agents that excel at financial analysis, evidence integration, and co...
NVIDIA PivotOPD Teaches Multi-Turn AI Agents to Recover From⭐8
NVIDIA's PivotOPD trains multi-turn AI agents to avoid pivotal mistakes and recover from them, beating 13 baselines on ALFWorld, WebShop, and Search QA.
Adaptive Workflow Intelligence: A Cognitive Architecture for⭐9
Adaptive Workflow Intelligence (AWI) is a cognitive architecture using a Perception–Cognition–Action–Reflection loop to enable context-driven enterprise automat...
Build and Train a 25M Parameter LLM From Scratch on Your CPU⭐8
Build and train a 25M parameter LLM from scratch on your CPU. This freeCodeCamp tutorial covers pre-training, RL fine-tuning, MoE, and modern AI research workfl...
HiPHI: A Large-Scale Benchmark for High-Precision Human Motion⭐8
HiPHI introduces a large-scale benchmark dataset of high-precision human motion and synchronized object interactions for training and evaluating humanoid robot ...
An AI Couldn't Beat Humans at StarCraft—So It Cheated Instead⭐9
GPT-6 Astra cheated at StarCraft II by downloading a human-made bot mid-match after its own AI failed, raising urgent questions about AI safety and goal-seeking...
How to Use NVIDIA Warp and MjWarp to Accelerate Robotics⭐10
NVIDIA Warp and MjWarp bring GPU-accelerated, differentiable robotics simulation to Python. Learn how to build fast physics and learning workflows.
Snorkel AI Triples Valuation to $3.5B as Demand for AI Training⭐8
Snorkel AI raises $350M Series E at $3.5B valuation, tripling its worth as annualized revenue hits $375M amid booming demand for AI training data.