“The use of fine-tuning for AI models has been declining as the capabilities of base models have improved, making it unnecessary for the vast majority of use cases.”