Workflow
科技部原副部长李萌:工程创新成为成就颠覆性创新更重要的形式
Di Yi Cai Jing Zi Xun·2025-06-27 10:25

Core Insights - DeepSeek has achieved a breakthrough in developing large models with lower costs while maintaining equivalent performance, prompting industry discussions on the efficiency revolution in large models [1] - Engineering innovation is seen as a crucial driver for disruptive innovation, with DeepSeek exemplifying the potential of engineering advancements in enhancing large model development [1][3] - The future of artificial intelligence will increasingly depend on the synergy between software and hardware, particularly in fields like humanoid robotics and advanced autonomous driving [1] Group 1 - The historical context of engineering innovation is highlighted, questioning why significant innovations often arise in specific locations, such as the steam engine revolution occurring in Manchester rather than London [3] - The interplay between theoretical breakthroughs and engineering optimizations is expected to lead future disruptive innovations, with both "0 to 1" and "1 to 100" processes being significant [3] - The efficiency revolution in large models is driven by a combination of architecture, strategy, and optimal software-hardware collaboration, indicating a shift from single-dimensional to multi-faceted understanding of innovation [3][4] Group 2 - DeepSeek's approach to developing large models emphasizes low computing power and cost while achieving performance equivalence, marking a shift in industry competition logic where efficiency is paramount for disruptive innovation [4] - The pursuit of energy efficiency is becoming increasingly important, suggesting that without high performance and energy efficiency, disruptive innovation may not occur [4] - Open-source initiatives are identified as essential for supporting the ecosystem of disruptive innovation [4] Group 3 - While focusing on disruptive innovation, it is crucial to consider potential disruptive harms, as current large model technologies exhibit incomplete explainability [5] - The governance of advanced AI technologies is becoming more urgent, especially as the reasoning capabilities of large models increase, leading to concerns about their compliance with instructions [5]