According to Vitalik Buterin, a high threshold for fine tuning AI models presents a double-edged sword. It allows developers more freedom to create fine-tuned models without being restricted. However, it also increases the liability of the original developers for outcomes that deviate significantly from the original model.