Technology cloud report original.
Whether to make a general large-scale model or a vertical large-scale model, this debate became more and more heated under the "Hundred Models War".
At present, major technology companies such as Microsoft, Google, Baidu, and Ali, etc., which focus on general-purpose large-scale models, have also begun to promote the commercialization of large-scale models in vertical fields.
For example, Microsoft and Google have integrated large-scale model technology into products such as operating systems, documents, search, and mail, demonstrating strong practical value.
The vertical large model is more direct in-depth specific industries and application scenarios, such as finance, medical care or retail, etc. Compared with the general large model, the vertical large model can more accurately meet the specific needs of the industry.
In the final analysis, whether it is a general-purpose or a vertical large-scale model, its core point is the landing scene and commercialization. The big model redefines the boundaries and possibilities of artificial intelligence, but what is more needed is to find a new living space and growth point in a specific business environment.
Among many vertical fields, the implementation of large models in the field of intelligent business BI is particularly eye-catching. As a key support tool for business decision-making, traditional BI has become inadequate in the era of big data.
The addition of vertical large models and the realization of conversational BI bring unprecedented possibilities to BI.
Data analysis is no longer a repertoire for specific professionals. Through conversational BI, every employee in an enterprise can interact with the system through natural language to obtain more intuitive and personalized insights.
From Traditional BI to Conversational BI
The first question to ask is, what is business intelligence (BI)?
Business Intelligence (BI) is a set of solutions that provide guidance and support for decision makers by analyzing, mining, integrating and displaying a large amount of unstructured data inside and outside the enterprise.
Specifically, BI is a complete set of data technology solutions consisting of data warehouses, query reports, data analysis, etc. Different business information systems integrate and summarize enterprise data.
BI can help companies gain a deep understanding of their own operations, monitor business processes, grasp market dynamics, and formulate strategies and tactics based on data analysis.
For example, BI can produce data visualization reports that satisfy different departments and employees at different levels, and can help front-line business personnel to implement operations such as business tracking, forecasting, and review; it can also help senior managers of enterprises to manage the cockpit through the business intelligence BI, Core KPI indicators, group Kanban, etc.
However, traditional BI systems are often complex and difficult to use, requiring professionals to operate and interpret them. The time delay and skill threshold have become obstacles to efficiency and popularization.
In addition, in the digital age, regardless of product development, marketing, financial management or customer service, using data to support decision-making in all aspects of business operations has become the daily routine of enterprises, and the demand for enterprise data analysis has increased significantly.
Similarly, enterprise managers, ordinary financial personnel and even business personnel need to use BI for more efficient data analysis and decision-making.
The emergence of conversational BI is precisely to solve these problems and needs.
Compared with traditional BI systems, conversational BI uses natural language processing technology, allowing users to query through natural language, just like talking with human analysts, so as to obtain the required information. This interactive method greatly reduces the barriers to use and improves efficiency.
**Popularity and convenience: **Conversational BI does not require professional skills and training, allowing more people to directly access and utilize enterprise data resources.
**Real-time and flexibility: **Conversational query allows users to obtain information in real time, quickly respond to temporary or urgent needs, and enhance the flexibility and response capabilities of enterprises.
**Personalization and intelligence: **Through intelligent large models, conversational BI can understand complex queries and provide customized answers based on individual needs and backgrounds.
**Integration and expansion capabilities: **Conversational BI can be more easily integrated with other systems, providing enterprises with a wider range of application scenarios and expansion possibilities.
The rise of conversational BI is not an accidental phenomenon, but an inevitable trend in the development of business intelligence. It reflects the urgent needs of modern enterprises for data analysis, as well as the pursuit of convenient, intelligent and efficient tools.
With continuous innovation and development in this field, conversational BI will become an important direction for future enterprise decision support.
credibility challenge
Although conversational BI improves efficiency, the credibility of answers to conversational BI based on general large models has become a problem.
ChatGPT-like products cannot provide completely accurate answers, mainly due to two points: first, ChatGPT-like products are better at processing natural language text data related tasks, and are not specially designed for data analysis; second, there may be fabricated facts in general-purpose large models , that is, "AI illusion".
Some products may have the so-called "AI illusion". In the field of BI, it may be a fabricated field, which may become a fatal problem in data analysis.
To solve this problem, some companies are actively looking for solutions. For example, Netease Shufan recently released Youshu ChatBI product, which emphasizes ensuring the credibility of data in four aspects: demand understanding, process verification, user intervention and product operation.
Yu Lihua, General Manager of Netease Shufan Big Data Product Line, believes that AI hallucinations are due to factors such as insufficient training data, encoding and decoding errors between text and representation, etc. The scary thing is that AI does not know that this is an illusion, and relying on Prompt alone ) is extremely difficult to correct.
In this regard, the core points of Netease Shufan to create a credible ChatBI are: understandable demand, verifiable process, user intervention, and product operation.
**Requirement understandable,** means that in order to make up for the threshold between user cognition and complex tools and improve accuracy, Netease Shufan uses the language understanding ability of large models to conduct demand analysis first, so that even if you don’t understand BI at all Users can also judge whether the data retrieval steps of the system are correct through the requirements analysis content.
**The process can be verified,** that is, the review can be realized with the help of the NL2SQL capability based on the large model. In order to improve NL2SQL capabilities, Netease Shufan customized and optimized more than 300,000 different types of queries and SQL.
Yu Lihua introduced that the NL2SQL domain model tuned by NetEase Shufan has reached the level of GPT-3.5.
**Users can intervene, **that is, the data model and query conditions are structured, and the user can intervene and adjust, switch the data model or query conditions.
**The product is operational,** that is, Netease Shufan has built a unique operational feedback mechanism for Youshu ChatBI products. Users can feedback the accuracy of the data, and administrators can operate the knowledge base, mark and optimize badcases, thereby improving data reliability.
Only relying on the NL2SQL model with GPT-3.5 strength in the analysis field cannot guarantee that the conversational BI will return 100% correct results. Product design is particularly important for implementation.
Yu Lihua said that the key to the credibility of ChatBI is to allow users to realize that it is wrong when the result is wrong, and to get the correct result through manual intervention.
In practical application, take the sales department scenario of a large supermarket chain as an example, the product personnel input "I want to see the monthly profit of North China in the first half of this year", and ChatBI can give corresponding results, and describe the query in natural language logic and steps.
Professionals who master SQL can click the "More" button to view the corresponding SQL. If the logic is wrong, for example, the user wants to see the data of the "order date" in the first half of the year, but the AI filters the "delivery date", the user can click "modify query conditions" to correct it.
According to NetEase Shufan, at present, there are several ChatBI applications in NetEase, which have covered non-technical personnel such as products, operations, marketing, and finance.
Find the balance between general and vertical large models
Returning to the discussion of generic and vertical large models at the beginning.
This time, a reporter threw this question to NetEase during the media communication session of NetEase Shufan City Tour.
Wang Yuan, vice president of NetEase, executive director of NetEase Hangzhou Research Institute, and general manager of NetEase Shufan, said: NetEase's current actual situation is to make both general-purpose large-scale models and vertical large-scale models. The final export is to make a vertical model, which is applied to the two most concerned areas of software development and data analysis from the perspective of the enterprise service market.
In order to achieve this goal, the NetEase team built a "public base" - a large model of NetEase "Yuyan". The model is jointly promoted by NetEase Fuxi Lab and Hangyan's artificial intelligence team, and more business teams will be added to build the base of the large model of NetEase Group in the later stage.
The exploration of BI products this time also achieved a balance between general and vertical models. Combining general and vertical models will help improve the accuracy and applicability of conversational BI, breaking the limitations of traditional business intelligence and opening up new possibilities.
In the new era of big models, many companies are standing at a crossroads, starting to find the balance between general big models and vertical industry big models.
The general model provides people with a wide range of application prospects with its powerful language understanding ability, while the vertical model for specific industries or needs can more accurately meet actual requirements.
In fact, general and vertical models do not exist in isolation, and the synergy between them may be the key to promoting the development of business intelligence to a higher level.
By precisely combining the advantages of both, companies can not only meet actual needs, but also lay a solid foundation for future innovation.
In the future, this balanced exploration will be more reflected in product design and technological innovation.
The general model may continue to expand its application in various fields, while the vertical model will focus more on certain key industries, such as finance, healthcare and education. The combination of the two will open new doors for smarter, more human-like interactions and services.
In the long run, the collaboration of general and vertical large models may drive the next milestone of artificial intelligence technology.
Is it possible to achieve the optimal solution for specific needs while maintaining generality? This is not only a technical issue, but also involves industrial strategy, business model and ethical considerations.
With the joint efforts of more companies and research institutions, finding the balance between general and vertical models will become the key to promoting continuous innovation and high-quality development of business intelligence.
This process will reveal how artificial intelligence can truly integrate into our daily life and work, bringing unprecedented convenience and opportunities to human beings.
epilogue
Conversational business intelligence represents a new direction in the field of BI, and its natural interaction and fast response time are changing the way enterprises analyze data. However, the accompanying credibility challenges will require an industry-wide effort to address.
In the future, through more R&D investment, cross-enterprise cooperation, and in-depth research on general and vertical models, we are expected to see more accurate, credible, and efficient conversational business intelligence products, driving the entire field of business analysis forward.
[About Science and Technology Cloud Report]
Focus on original enterprise-level content experts - technology cloud reports. Founded in 2015, it is the top 10 media in the cutting-edge enterprise IT field. Recognized by the Ministry of Industry and Information Technology, Trusted Cloud, one of the official media designated by the Global Cloud Computing Conference. In-depth original reports on cloud computing, big data, artificial intelligence, blockchain and other fields.