Reasoning trace: learning primality from 100 examples

GPT-5-Thinking progressively refines its hypothesis from simple digit heuristics to a full Miller–Rabin test.