Group 1 - The concept of Physical AI was proposed in 2020 and is seen as a parallel to the digital world, with NVIDIA's CEO Jensen Huang defining it as a significant direction for AI development [1] - The industry is eager to create a killer application for Physical AI similar to ChatGPT, with discussions on when this "ChatGPT moment" will occur and the existing bottlenecks [1] - Optimistic predictions suggest that achieving a 90-95% success rate in 100-200 common household tasks could signify the arrival of Physical AI's "ChatGPT moment," potentially within 2-3 years [1] Group 2 - Wang Yequan, head of the FLM team at the Zhiyuan Research Institute, provides a longer-term estimate for reaching a ChatGPT-like level, suggesting it may take 5-10 years due to the current stage of embodied intelligence being around GPT-1.6 or 1.7 [2] - Despite differing timelines, experts agree that Physical AI faces significant challenges, particularly in data collection and modeling [2] - The difficulty in data collection for Physical AI arises from the need for frequent interactions with the physical world, making it harder to gather comprehensive datasets compared to digital models [2] Group 3 - Zhu Zheng believes that world models could address data scarcity by generating synthetic data to match the volume and generalization of internet data in the digital realm [3] - Wang Yequan identifies modeling methods as the primary challenge for Physical AI, emphasizing the need for large-scale self-supervised data to create a foundational model, which requires substantial resources [3] - The proposed approach involves aligning and reinforcing the model's capabilities for real-world applications, but the lack of suitable modeling methods and data organization remains a significant hurdle [3]