The landscape of AI is becoming increasingly consolidated, with high capital expenditure required for training large language models, limiting this capability to a few organizations. However, the emergence of frontier-level open source models is noteworthy, as they offer flexibility and the ability to deeply experiment with AI technologies. This new wave of open source AI allows developers to adapt and manipulate models in a way that's not possible with API-accessible models. Technologies such as 'task vectors' enable real-time modifications of model weights, facilitating innovative applications. As smaller models gain efficiency and decrease in cost, their utility is being validated, emphasizing user control and adaptability. This trend towards smaller, locally-run models is particularly advantageous for edge device deployment, where network latency can be a concern. The duality of large and small models in the AI ecosystem presents exciting possibilities for hobby projects and practical applications alike.