“Asha Sharma predicts that in the future, more money will be spent on post-training AI models (fine-tuning, RL) than on pre-training them.”