“Ari Morkos asserts that simply scaling up compute and data for AI model training is now hitting diminishing returns, an outcome that was predicted by original scaling laws research.”