AI systems are slowly morphing into neurosymoblic AI o1 uses CoT RL with reward model, not just pure deep learning AlphaGeometry and AlphaProof uses LLM with symbolic Lean AlphaCode uses MCTS and sampling etc.