The discussion highlights tools and resources available for training AI models on a budget using Google Colab. Notebooks created for GRPO (Guided Reinforcement Policy Optimization) and general finetuning of the Llama 3.1 8B model were shared, which enable users without deep pockets to experiment with model development. The community seems enthusiastic about the surge in competition spurred by open architectures, fostering innovation. Users discuss the feasibility of achieving greater advancements if (1) training datasets were openly accessible and (2) copyright constraints were relaxed, while others express concerns about the accuracy of the post title which suggests starting from scratch when it actually builds on pre-existing models. Furthermore, some users express frustration over the challenges of using AI for creative purposes, specifically in music production, highlighting a gap in user-friendly tools for such applications.