
The AlphaGo 3-3 analogy assumes verification symmetry that's missing here. AlphaGo's 3-3 invasion had MCTS rollouts + win probability as a built-in quality signal, so once humans saw the move they could verify it was correct without understanding why. A single Erdos proof lacks the equivalent signal, whether GPT-5.4's solution is novel or stitched together from existing math literature is not separable without a formal verification pipeline. The AlphaGo-moment framing needs reproducibility of the discovery pattern on 10-20 unsolved problems to hold.
English





























