Gary Marcus質疑超算規模並倡導神經符號AI
gary marcus反對對超算規模的巨額押注，敦促轉向神經符號ai和世界模型以實現可靠性。
through interviews at web summit and bug bash 2026, marcus highlights the risks of current generative ai strategies. he emphasizes that software verification remains critical in the llm era and suggests that blending neural networks with symbolic logic offers superior reliability. marcus also referenced the emerging potential of autonomous agents in consumer applications.