Core Insights - The rapid development of large model technology has made top AI talent a key target for tech companies [1] - JD.com has launched the TGT program to recruit top technical talent globally, focusing on various advanced fields [1] - JD.com aims to enhance its AI infrastructure through collaboration with domestic hardware manufacturers and academic institutions [1][2] Group 1: Talent Recruitment and Development - JD.com introduced the TGT program targeting global students and recent graduates in eight technical fields, covering over 80 advanced topics [1] - The talent structure is pyramid-shaped, with top leaders guiding the direction and mid-level technical staff being recruited through the TGT program [1] - JD.com plans to offer 18,000 core positions for the 2025 graduating class [1] Group 2: AI Infrastructure and Optimization - JD.com is addressing hardware efficiency and stability issues through software optimization, achieving a 30%-40% performance improvement with its self-developed xLLM model [2] - The company emphasizes real-world application of algorithms in retail, focusing on solving business problems in high-demand scenarios [2] - JD.com is advancing large model applications through restructuring traditional processes, integrating generative models, and exploring native applications [2] Group 3: AI Application and User Experience - During the 618 shopping festival, the usage of large models increased by 130% compared to last year's Double 11 event, with over 14,000 AI agents operating internally [3] - The transition from keyword matching to semantic understanding and multimodal interaction is being driven by large models [3] - The proportion of new graduates involved in large model application projects is significantly higher than in traditional business areas [3]

Core Insights - JD's 618 shopping festival in 2025 saw over 100% year-on-year growth in the number of users placing orders, with total orders exceeding 2.2 billion [2] - AI-related products experienced significant sales growth, with AI laptops up 151%, AI tablets up 149%, and AI learning hardware up 120% [2] - The use of AI models increased by 130% compared to the previous year's Double Eleven event, with over 14,000 AI agents operating internally [2][3] AI Applications and Developments - JD has been utilizing AI technology since 2012, evolving from traditional methods to advanced NLP and deep learning techniques, achieving over 90% accuracy in AI applications [5] - The company aims to serve over 50,000 brand merchants with AI digital humans by the end of 2025, focusing on making AI technology more accessible [3] - JD's AI applications span retail, logistics, health, and industrial sectors, enhancing operational efficiency and reducing costs [6][7] Performance Metrics - During the 618 event, JD's AI-driven operations improved conversion rates in live streaming by 30%, with AI digital humans costing only one-tenth of human hosts [11] - The AI customer service system handled over 2.5 billion inquiries during the event, showcasing the scale of AI integration in customer interactions [11] - JD's logistics operations utilized AI models over 1.3 billion times, benefiting more than 500,000 frontline logistics employees [11] Future Directions - JD is exploring "embodied intelligence" technologies, integrating AI models with various hardware to enhance user interaction [12] - The company is focused on developing AGI through three main directions: large language models, multimodal capabilities, and embodied intelligence [12] - JD's industrial AI model, JoyIndustrial, aims for significant cost reductions in supply chain management, achieving a 99.3% reduction in technical costs [11]