news

More than a year after SF Express President Wang Wei publicly stated that "SF Express has begun to study ChatGPT", SF Express has released new news on the big model.

On August 19, a reporter from the "Daily Economic News" learned from SF Express that SF Technology launched its independently developed "Fengzhi" logistics decision-making model on August 18. The model aims to apply big model technology to intelligent analysis of logistics supply chains, sales forecasting, transportation route optimization, packaging optimization and other decision-making areas.

As early as last year when ChatGPT emerged and triggered global discussions, Wang Wei publicly expressed his interest in and sense of crisis about this technology. However, compared with other companies' explorations in the field of logistics big models, SF Express still seems to be late.

Since last year, JD.com and Cainiao have both released big models in the logistics industry. JD Logistics disclosed that in the first half of this year, JD Logistics applied the logistics big model to JD.com's "618" security work for the first time. As of the end of June, the logistics big model has been called more than 30 million times in total, empowering hundreds of thousands of front-line employees. In March this year, Huolala CTO Zhang Hao publicly revealed that Huolala's self-developed freight worry-free big model has been applied in multiple scenarios, with a question-and-answer accuracy rate of more than 90%.

Now with the entry of SF Express, the melee of logistics big models has escalated again. In the competition among logistics companies in the big model era, who will seize the initiative?

SF Express launches logistics decision-making big model as the first step out of big model layout

"Reality is always cruel. Let me tell you the truth. 'Dick Online' (SF Express's 30th anniversary speech) was not written by me, but by a big model." In the first half of last year, Wang Wei shared a screenshot of a conversation.

"In the process of rapid technological development, SF Express cannot be lazy and needs to rely on new things and new tools to keep up with the times." Wang Wei said on the internal platform, and bluntly stated, "In this era, we are all being eaten by the tigers of technology or innovation in minutes, rather than being killed by our peers."

Now SF Express has finally launched its own large model product. The reporter of Daily Economic News learned that in order to solve the problem of implementing large models in supply chain operation scenarios, SF Technology has integrated the interactive advantages of large language models with the professional depth of traditional small models to build a supply chain intelligent body. This intelligent body is based on SF Technology's Fengzhiyun ecosystem, and can accurately inform customers of the cause of the problem when their sales fluctuate, provide decision-making basis for managers, and help them choose appropriate response strategies.

SF Technology told the reporter of the Daily Economic News that the "Fengzhi" logistics decision-making model has built a multi-level and multi-channel demand forecasting model based on the multi-modal large model capabilities to achieve more accurate forecasting results. Not only that, the change in calculation method has also reduced the number of model requirements and resource consumption, and improved resource utilization efficiency.

Taking a practical case as an example, its server resource requirements were greatly reduced, the running time efficiency was increased by 120 times, and the prediction accuracy was increased by 5%. This improvement is very difficult to achieve with traditional models, and it also marks that the "Fengzhi" logistics decision-making model has achieved a major technological breakthrough in the field of supply chain demand forecasting.

However, compared with logistics companies that released large-scale model products as early as last year, SF Express's move is not considered early.

"In recent years, general artificial intelligence, especially large language model technology, has made significant progress, profoundly changing the ecology of many industries. However, in the complex and highly specialized field of logistics supply chain, its application potential has not yet been fully explored and released," said Tang Kai, vice president of SF Technology.

Tang Kai also mentioned that supply chain operation is a highly professional and rigorous field. Although the current large model technology shows great potential, its in-depth application in this field is limited to a certain extent by problems such as accurate calculation and avoiding "hallucinations".

Tang Kai further stated that the key point is that the results of supply chain operation decisions and analysis must be traceable, traceable and verifiable, which is the so-called "white box" feature. This is fundamentally different from the "black box" output method commonly used in current large models. This core difference has become the main challenge faced by large models in the widespread promotion of the supply chain field, hindering the full display of their capabilities.

It is worth mentioning that on the same day (August 18) when SF Technology launched the “Fengzhi” logistics decision-making big model, SF Technology also reached ecological cooperation with a number of big model startups including Zero One Everything. SF’s further actions and layout in the field of big models have aroused expectations in the industry.

The giants are here! The big model has been explored and implemented in many logistics scenarios. Who can seize the opportunity?

The industry believes that among many industrial scenarios, road freight has the characteristics of a complete system, many links, complex data, and rich scenarios, and is one of the main areas for the implementation of big models. The in-depth application of big model technology is expected to bring about tremendous changes in the operational efficiency, service experience, and business innovation of the entire logistics chain.

Therefore, since last year, Cainiao, JD Logistics and others have already taken action to release large-scale model products for the logistics supply chain field. In June last year, Cainiao Supply Chain released a digital supply chain product "Tianji Pi" based on a large model. Through Cainiao algorithm + generative AI-assisted decision-making based on large models, it has achieved quality and efficiency improvements in sales forecasting, replenishment planning and inventory health, and has effectively promoted the logistics supply chain into the era of large models.

On July 13 last year, at the 2023 JD Global Technology Explorers Conference, JD officially launched the Yanxi Big Model, which supports language, voice, vision, multimodality, etc., integrating 70% general data and 30% digital supply chain native data, and is committed to deepening logistics industry scenarios and solving real pain points.

Baidu Maps also announced at the end of September last year that based on Baidu's big model technology capabilities and combined with the characteristics of the logistics industry, it officially launched the Beta version of the logistics big model, taking the lead in applying it in the two major fields of logistics address resolution and logistics scheduling decisions. In October last year, Fuyou Truck and Tencent held a signing ceremony to reach a comprehensive strategic cooperation around the creation of the first digital freight big model.

In March this year, Huolala's self-developed freight worry-free large model focused on scenario-based and lightweight, positioning itself as "the logistics expert by your side". At present, the accuracy rate of its freight factual questions and answers exceeds 90%. In the same month, the "Logistics Intelligence Alliance", the first alliance in the logistics industry focusing on large-scale model application research and practice, was established. The alliance was jointly established by Alibaba Cloud, Cainiao, AutoNavi, COSCO Shipping, China Eastern Airlines Logistics, YTO Express, STO Express, ZTO Express, Deppon Express, G7 Easy Flow, Ground Metro, and Zhejiang University Intelligent Transportation Research Institute at the 2024 Digital Logistics Summit.

Now the launch of SF Express’ self-developed big model also means that all logistics giants have arrived, and a more intense war has officially begun in the field of logistics big models.

However, Chen Guanling, technical partner of Fuyou Truck, once told a reporter from the "Daily Economic News" that the current logistics big model is still in a very early stage. In a vertical industry like logistics, because the products made are not creative products, the accuracy of the big model output is very high. The current big model is not accurate enough and has a strong randomness. More algorithmic improvements and a large amount of data and scene enrichment are needed to solve this problem. This is also the biggest challenge facing the current logistics big model.

The industry consensus is that 2024 is the year when generative AI will be integrated with the industry and truly implemented. Now, with more and more important guests from the logistics industry entering the market, who can seize the initiative in the logistics field has become the biggest highlight and suspense in the future.

Daily Economic News