Workflow
VLA(视觉语言行动模型)
icon
Search documents
专家:2035年机器人数量或比人多
Core Insights - The rapid development of the AI industry is accelerating iterations across various sectors, presenting significant industrial opportunities [1] Group 1: Trends in AI Industry - The first major trend is the transition from discriminative AI to generative AI, now evolving towards agent-based AI, with task length doubling and accuracy exceeding 50% in the past seven months [3] - The second trend indicates a slowdown in the scaling law during the pre-training phase, shifting focus to post-training stages like inference and agent applications, with inference costs decreasing by 10 times while computational complexity for agents has increased by 10 times [3] - The third trend highlights the rapid development of physical and biological intelligence, particularly in the smart driving sector, predicting that by 2030, 10% of vehicles will possess Level 4 autonomous driving capabilities [3] Group 2: Future Projections and Risks - The fourth trend points to a significant rise in AI risks, with the emergence of agents increasing risks at least twofold, necessitating greater attention from global enterprises and governments [4] - The fifth trend reveals a new industrial landscape for AI, characterized by a combination of foundational large models, vertical models, and edge models, with expectations that by 2026, there will be approximately 8-10 foundational large models globally, including 3-4 from China and 3-4 from the U.S. [4] - The future is expected to favor open-source models, with a projected ratio of 4:1 between open-source and closed-source models [4]
理想汽车推送OTA 8.0版本,李想称公司辅助驾驶开始“全面领先”,VLA优于世界模型?
Mei Ri Jing Ji Xin Wen· 2025-09-12 10:06
Core Viewpoint - Li Auto's advanced driver assistance and smart cockpit have transitioned from "partially leading" to "fully leading" following the OTA 8.0 update of their vehicle system [1] Group 1: OTA 8.0 Update - The OTA 8.0 version has officially launched, enhancing driver assistance, smart cockpit, and smart electric features [3] - The new VLA (Vision-Language-Action Model) driver model is being fully pushed to Li MEGA and L series AD Max models [3] - Li Auto's chairman, Li Xiang, described VLA as the third generation of their driver assistance technology, emphasizing its ability to understand road conditions, comprehend human commands, and remember user habits [3] Group 2: VLA Model Features - The current version of VLA is referred to as a "crippled version" due to the temporary absence of a highly praised feature [4] - Li Auto has acknowledged the need for a cautious approach in rolling out new features, especially after the suspension of the VLA remote summon function [4] - The VLA model enhances the accuracy of route selection in complex scenarios and remembers user speed preferences for specific roads [6] Group 3: Industry Competition and Technology - Other companies like Yuanrong Qixing and XPeng Motors are also developing VLA models, indicating a competitive landscape in this technology [7] - The VLA model is seen as an "intelligent enhanced version" of end-to-end models, addressing challenges in handling unseen scenarios [8] - The VLA model integrates perception, action execution, and language processing, enhancing its ability to understand and make decisions in complex environments [8] Group 4: Differing Approaches - Huawei's approach focuses on the World Action model, which bypasses the language processing step, emphasizing direct control through vision [12] - The debate between VLA and world models highlights differing strategies in achieving advanced autonomous driving capabilities [12][13] - Experts suggest that both VLA and world models can coexist and complement each other, with different companies choosing paths based on their specific goals [13]