sf express's fengyu language model is here! its logistics vertical capabilities surpass general models and have been applied in more than 20 scenarios
2024-09-10
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
on september 8, sf technology released the vertical language model of the logistics industry, fengyu, at the shenzhen international artificial intelligence exhibition, and demonstrated the practical application of the big model in more than 20 scenarios in sf's marketing, customer service, collection and delivery, international customs and other business sectors. at the press conference, yao jianquan, an academician of the chinese academy of sciences, pan yi, an academician of the american institute of medical and biological engineering, lan zhihua, deputy director of the shenzhen postal administration, geng yankun, cio of sf group and ceo of sf technology, and other relevant experts and leaders witnessed this milestone moment.
[photo: fengyu large model release ceremony]
this model is currently the most targeted large language model for the pain points of the logistics industry, and its application scenarios in the logistics industry are extremely rich. it can efficiently empower various positions in the logistics industry, allowing everyone to quickly become experienced job experts. its application effects related to logistics scenarios and logistics vertical domain capabilities have been greatly improved relative to the general base. at the same time, the model fully considers the actual situation of the logistics industry, achieves a balance between effect and use cost, and truly allows every business to afford the large model for empowerment, achieving the best balance of professionalism, reliability, and cost.
at the launch ceremony, song xiang, vice president of sf technology aiot, said that the logistics industry is a service industry, and the knowledge and experience reserves of relevant practitioners are closely related to service quality and operational efficiency. only by empowering each position based on a large model can everyone quickly become an experienced position expert, complete the judgment of related operations faster and more accurately, gain insight into potential business opportunities, and provide better services to customers.
at the same time, he further introduced the application effects of sf express's big model in business segments such as marketing, customer service, collection and delivery, and international customs.
for customers, in terms of accelerating the conversion of business opportunities, sf express currently applies it to the output of marketing graphics and text-related content, product and service recommendations for personalized needs, return order identification and risk control, etc. in terms of customer service, sf express can not only help customer service quickly extract key information from customer conversations and form service summaries through big models, but also gain insight into opportunities in customer feedback and further improve service quality. at present, the accuracy of summaries based on big models has exceeded 95%, which has reduced the average processing time of customer service personnel after conversations with customers by 30%.
for the delivery guys, sf express uses digital humans to provide one-on-one training for new employees from the moment they join the company. the big model can also be used to ask various questions during daily collection and delivery operations, such as whether the items can be sent, whether insurance is recommended, how to package them, how much products with different time limits cost, etc. the accuracy rate of locating the delivery guys' questions is over 98%, and each conversation can save the delivery guys 3 minutes on average. at the same time, the big model will also analyze the delivery guys' feedback, adjust the delivery guys' service strategies in a timely manner, improve the efficiency of handling key issues, and make the delivery guys more satisfied.
at the same time, sf express has achieved comprehensive and inclusive application of large models. for example, office assistants, ai agents, etc., have covered more than 1,000 positions. it also provides a one-stop intelligent body code-free development platform, allowing more different business organizations to build their own large model question-and-answer robots within 1 day according to their own needs, realizing comprehensive empowerment of sf express's various businesses.
the application of large models in the customs scene is rare in the industry. when sf express uses large models, it can quickly determine whether the items meet the customs clearance standards of relevant countries when faced with multi-national terms and various languages. at the same time, it can standardize the names of declared goods according to relevant requirements, and can also quickly determine whether the relevant sent items can pass customs through high-speed scanners, helping customers and relevant staff to make judgments while improving customs clearance efficiency.
in addition, in the afternoon sub-forum, jiang shengpei, technical director of sf technology's big model, further introduced fengyu's thinking on technology research and development. he said that sf has developed a large language model for the vertical field of the industry with a balanced approach of comprehensive consideration of effect and cost of use. in terms of training data, about 20% of fengyu's training data is vertical data related to the logistics supply chain of sf and the industry, and through continued pre-training, supervised fine-tuning, and large model training methods based on artificial feedback reinforcement learning, fengyu's big model has become more logistics-savvy and more reliable. at the same time, sf has also established an evaluation system for large models in the logistics field, and conducted a detailed evaluation of the general capabilities and logistics vertical capabilities of the model. at present, fengyu has achieved a comprehensive surpassing of larger-size general models in the logistics vertical field with smaller-size models. for example, in the brother's question-and-answer scenario, fengyu's error rate has been reduced by 58% compared with the original general base; in the customer service form filling scenario, the error rate of real-time logistics information extraction has been reduced by 52%. this fully reflects the significance and value of the vertical domain large model.
in the context of the country's promotion of new quality productivity, artificial intelligence has become one of the core driving forces of economic development. to truly bring out the value of big models, we also need to focus on solving specific problems in the industry and achieve industrial breakthroughs. in the future, with the further development and maturity of big model technology, sf express believes that the application of big models in the logistics industry will be more extensive and in-depth, allowing everyone to enjoy faster and more intimate logistics services enabled by big models. (china economic news network)
