This session is for blockchain developers and researchers seeking to understand the realities and challenges of integrating machine learning with blockchain technology. We will debunk common misconceptions, such as "onchain machine learning" is not typically executed on the blockchain. The talk will address the frequent concerns about non-determinism in ML models and their compatibility with blockchain systems. Importantly, we'll clarify that "onchain" ML does not inherently guarantee verifiability. Instead, we'll explore cutting-edge techniques for achieving verifiable machine learning in blockchain contexts, including optimistic systems, zero-knowledge proofs, and trusted execution environments. By demystifying these concepts, attendees will gain valuable insights into the current state of the art, enabling them to make informed decisions when designing and implementing ML-powered blockchain applications.