AI碰到天花板？地平线苏菁再“开麦”：智驾苦日子又要来了

Core Insights - The current generation of deep learning technology may be reaching a bottleneck, leading to a phase of optimization rather than fundamental theoretical breakthroughs in autonomous driving over the next three years [1][3] - The transition from rule-based to data-driven paradigms in autonomous driving is exemplified by Tesla's FSD V12, which integrates perception, decision-making, and control into a single neural network model [2] - The industry is expected to see significant advancements in L2 level assisted driving, with urban driving assistance becoming more common in vehicles priced around 100,000 yuan [2] Group 1 - The sentiment in the autonomous driving industry is mixed, with some experts expressing skepticism about the future potential of AI and AGI in the next three to five years [3] - The cost of developing and testing end-to-end systems is extremely high, with estimates suggesting that a single round of testing could cost around 1 billion yuan, highlighting the financial risks involved [3] Group 2 - The adoption of end-to-end technology in the autonomous driving sector is anticipated to unify methodologies for L2 and L4 levels, enhancing the driving experience while reducing deployment costs [2] - The shift towards more human-like driving systems is expected to create a significant growth period for L2 level assisted driving technologies [2]