Show HN: Hands-on course for building RL environments for LLMs (github.com)

A new hands-on course designed to explore Reinforcement Learning (RL) environments for training and evaluating language models (LLMs) has been announced, focusing on creating an interactive Tic Tac Toe environment. This innovative approach moves beyond traditional supervised fine-tuning methods, allowing models to learn from exploration and adapt in ways that curated datasets cannot facilitate. The course is aimed at AI engineers with knowledge of LLMs, experienced RL practitioners, and curious individuals interested in the intersection of reasoning models and RL post-training. Participants will build a Tic Tac Toe environment using an open-source library called Verifiers, which serves as a framework for developing RL environments. The curriculum covers essential topics such as evaluating existing models, generating synthetic data for supervised fine-tuning, and implementing RL training to enhance model performance, ultimately transforming a small language model (LiquidAI/LFM2-2.6B) into a competitor capable of beating more advanced models like gpt-5-mini. This course not only provides practical insights into RL applications in LLMs but also fosters community collaboration through hands-on experimentation and open feedback channels on GitHub.