Apple study finds "a fundamental scaling limitation" in LLM reasoning modelsthe-decoder.com9 pointsrbliona year ago