with LLMs able to generate infinite synthetic data to train on it seems like AGI is just around the corner