Seven years after Apache Kylin commercialization, Han Qing no longer publicly talks about open source commercialization, because open source has long become Kyligence's "engine." When he saw a revolutionary change in the way humans and computers interacted, he realized that he could no longer look at the problems in the data field through the eyes of the past. At the critical moment when AI "invaded" the field of big data, he had a new direction to fight.
Interviewees:
Han Qing, co-founder and CEO of Kyligence, co-founder and PMC member of Apache Kylin, leader of open source big data OLAP, the first VP of the top project of the Apache Software Foundation in China, technical director of Microsoft community, Tencent Cloud Most Valuable Expert (TVP), director and think tank expert of Fintech Innovation Alliance, member of the Shanghai Big Data Alliance, member of the Digital Finance Working Committee of the Internet Society of China.
Author | He Miao
Listing | CSDN（ID：CSDNnews）
Han Qing, one of the earliest "crab-eaters" in the commercialization of China's open source projects. There is a momentum to strive for the upper echelon in his bones, and on the road of embracing AI, he believes that "if you don't innovate, you will be eliminated, and if you don't do it, you will fall behind."
As the commercial company behind Apache Kylin, China's first Apache top open source project, Kyligence has gone steadily and quickly on the road to commercialization. Thanks to Han Qing's daring to think and do, there is also a good partner who can help him realize his dream - co-founder and CTO Li Yang. As the two pillars of Kyligence, they have led the company to a gallop in seven years to become the backbone of the technology behind industries such as finance, manufacturing, retail, and pharmaceuticals. At the important turning point of AI reshaping all walks of life, it took the lead in landing AI applications in the field of big data.
Left Han Qing, right Li Yang
According to Han Qing, Kyligence Copilot AI digital intelligence assistant took less than half a year from conception to implementation. Is it because it rolls hard enough? No, it's more of a natural transformation that has accumulated over time, and quantitative changes produce qualitative changes. He has been working hard to democratize specialized tools, and the AI wave has made this goal a more comfortable outlet.
What are the opportunities and challenges of combining large models and data applications? Where will Han Qing, who has the OLAP engine + indicator platform + Copilot "troika", lead Kyligence towards? In this issue, Qing Han, co-founder and CEO of Kyligence, details the opportunities and challenges of big data embracing AI, and how an entrepreneur who is still full of enthusiasm can start his own transformation path.
AI "invades" big data and keeps the theme of democratizing specialized tools
"Near Craftsman": What changes has the big model of hot development in the first half of the year brought to the big data industry? You have been in the industry for many years, what do you think is constant?
Han Qing: The biggest change brought about by this wave of AI is the change in the mode of human-computer interaction. In the past, people relied on professionals to use data, managers wanted to see indicators to entrust professional programmers or engineers, and the interaction mode was layer-by-layer subcontracting. Today, with AI as an amplifier, many new efficiency tools have been born, and in the field of data, this is also a revolutionary change.
Natural language interaction technology has been developed for decades, and it has reached the moment when quantitative change leads to qualitative change, behind which is the accumulation of technology. On the one hand, the cost of computing power is declining, and on the other hand, large model algorithms are not complex in nature. Data + computing power, vigorous can produce miracles.
Our insistence on "simplifying complex data problems and making professional and complex tools popular and simplified" has not changed. At the same time, the overall operation thinking of all enterprises and the essence of managing the company have not changed, and it is still to improve efficiency and reduce costs.
Near Craftsman: What are your new ideas in the face of this wave of AI revolution?
Han Qing: When I saw the revolutionary changes in human-computer interaction, I realized that I could no longer look at the problems in the data field with the eyes of the past. I've been wondering how AI can change the industry. Over the years, we've been doing data services, what chemistry will be generated by combining standardized products with new interactions? That's interesting, and then step by step there is Kyligence Copilot AI Digital Intelligence Assistant (preview).
Democratizing specialized tools is our consistent approach. OLAP engine is a very professional tool, a small number of enterprises can use, we based on this to make an indicator platform, most business personnel can also understand, can use, and now AI data assistant, almost everyone can use.
Near Smith: Will Kyligence make big models? How does the AI Digital Intelligence Assistant work?
Han Qing: We will not make big models, but AI digital intelligence assistant is an application based on large models. When many people pay attention to large models, I think the most important thing is actually how to do the application, in the field of data and analysis, we may be the earliest commercial application. A brief introduction to its technical principles.
The underlying large model currently docked by AI Digital Intelligence Assistant comes from OpenAI and some open source large models, considering the domestic use specifications, in the actual enterprise access, there are generally two choices: based on the open source basic model to do secondary training, or purchase domestic model suppliers, on this basis to do secondary training. At present, we have supported access to the enterprise's own (private) large model, or only ten lines of code to embed Copilot into the enterprise's own application, which is very convenient.
Generally, there are many uncertainties in the answering process of the dialogue large model, but the working logic of the AI digital intelligence assistant is different from the traditional language model. Its instruction execution is answered data questions on a controlled indicator platform, not generated using language models, which eliminates many uncertainties and security risks. The answer to the question also cannot go beyond permissions to obtain data, and from this point of view, the user's original data is safe.
Where do the six major difficulties in AI application landing break through?
"Near Craftsman": The needs of pioneer enterprises are the largest and most complex, but many people will always have doubts about handing over all data to AI, what problems may exist in the process of AI applications?
Han Qing: In March this year, the prototype of the AI digital intelligence assistant was born, but it has not moved, because it has never figured out what it can do and what effect it will have, so I hope to find a breakthrough and do my best. After talking to users over the past few months, it has a new twist. Users are actually more anxious, everyone knows that AI will definitely change the industry, if the industry pioneers do not use AI first, they will fall behind. And they, cannot lag behind.
However, in the actual landing, the following difficulties will still be encountered:
First, compliance. Compliance is very important. Today, the big models of Open AI can't be used by banks.
Second, security. Whether the company's operational indicators and all data can be released to AI to process, there is still too much uncertainty, which is a problem for the entire industry.
Third, the problem of internal culture. Every company has its own data culture, and generally startups have relatively flat management teams and relaxed atmospheres, but state-owned enterprises may be a different culture. But that's where AI has the opportunity to change industries.
Fourth, problem alignment. How to make AI understand your true intentions is also an industry problem. Human expression is often duplicitous or unspoken, which leads to external behavior patterns that may be contrary to their own intentions, which is also a major difficulty for machines to deal with.
Fifth, how to get employees to use AI tools well. Different ways to ask the same question may produce different results, so training in questioning methods is also required.
Sixth, what if enterprises want to use it, but do not have data accumulation? Data maturity determines how effective AI digital intelligence assistants can be. It is difficult for a smart woman to cook without rice, and enterprises with a better data foundation will feel better when using it.
In addition, the maturity of domestic domestic models is also a major factor, and it is expected to have better development.
"Near Craftsman": You emphasized that Kyligence has been lowering the threshold for using user data, but you also mentioned that enterprises must have a certain data foundation if they want to use AI digital intelligence assistants well, which is a relatively high threshold. How to understand the contradictions?
Han Qing: This is actually what we often call the concept of "using and governing at the same time". In the past, the threshold for the use of data was high because enterprises needed to govern first and then use, data caliber to be aligned, data to be aligned, data to be aligned, etc., and governance required a lot of manual input, but now AI can replace this part. At the same time, our indicator platform can assist different companies to build some common indicators, after one or two months of use, after running through, the commonality of enterprises begins to appear, and then through the AI digital intelligence assistant unified governance feedback to users, the industry model can be gradually established, which is the transformation process of AI to the software industry.
"Near Craftsman": You mentioned that enterprises are very concerned about security issues, what technical means can ensure the security of users' data in the application process? Is it privacy computing, or is it isolation of computational storage? How to assign and implement users' data usage rights?
Han Qing: To put it simply, Kyligence cannot interfere with user data privacy, and with our SaaS service, the integrated computing of AI digital intelligence assistant and the user's office system are completely isolated. We access the system through a designated springboard machine, which also goes through an approval process, and all actions are recorded for review, which is the most basic structure.
The working principle of AI digital intelligence assistant is also well understood, it receives an indicator query instruction - for example, when executing the search sales system, it will also retrieve the permissions of the initiator of the instruction, and the user cannot obtain data and query results outside his own authority.
Technically, it is not difficult to implement data rights management for users. There is data control based on the role dimension within the enterprise, for example, Xiao Zhang can see the indicators of the department, but Xiao Li cannot see it; There is also a dimension of data strength, for example, sales in Shanghai can only see the data of Shanghai, and cannot see other regions.
To fully solve the concerns of enterprises, the most important thing is to have international authoritative certification, Kyligence products have passed SOC2 Type 1, Type 2, ISO9001, ISO27001 and other certifications and audits, which is the world's most important data security certification, our high security, high confidentiality and high availability is guaranteed.
Where is the "OLAP Engine + Indicator Platform + Copilot" troika going?
Kyligence: As a commercial company that started with open source, why is there less talk about open source these days?
Han Qing: At the beginning of our business, our open source community has done well, and then tried to commercialize it, and has been exploring its own transformation path. Nowadays, there is little talk about open source commercialization, on the one hand, to meet market demand, and on the other hand, because it is doing enterprise services. Open source is our important engine, if we compare it to selling cars, it can be said that we are now selling "complete vehicles", and "engines" are provided by open source. Selling complete vehicles and 4S shop insurance, this part is called service.
"Near Craftsman": When developing three different series of products at the same time, how to measure the input and output of enterprises?
Han Qing: Startups should always find a balance between "doing the original thing" and "exploring innovation". Just like the mainland's strategy in military affairs, it will always be "service generation, development generation, exploration generation". For us, it's hard to say what the account will be, but what is clear to me is that if you don't innovate, you will be eliminated, and if you don't do it, you will fall behind. If a startup sticks to the rules and does not move forward, it will inevitably fall. The same is true for our users, pursuing new technologies, experimenting with new scenarios, and exploring ideas with each other.
Near Smith: What is the greatest value of Kyligence Copilot AI?
Qing Han: It allowed Kyligence to start from a technology company to a real management software company. We used to provide tools, now we provide platforms and even management methodologies, which is something we have always wanted to do. Being able to define industry standards, define the future of the industry, and even lead industry trends is what startups or companies deep in the industry should do.
Near Craftsman: What are the challenges of this transformation?
Han Qing: From a technology software company to a management software company, for us with technical backgrounds, the next difficulty is that management knowledge needs to be practiced and precipitated.
The second challenge is that China's corporate management is relatively extensive, and there are still big differences from companies in the United States and Germany. A large number of Chinese companies pay attention to "rule by people", while Western companies do better in process management. How to do a good job of refined management and compliance? We can't change the method of enterprise management, so making a good platform to improve the efficiency and results of corporate governance is what we want to do.
Near Craftsman: In Kyligence's business logic, focus on customers, not competitors and markets, why?
Han Qing: I have always emphasized internally: "Focus on your customers not your competitors." This phrase is very popular in the United States. There is competition in the industry, don't be anxious, focusing on competitors will only make you worse and worse, focus on customers and they will teach you.
To maintain its competitiveness, two points are important: First, the uniqueness and differentiation of the product. Second, the establishment of barriers. Nowadays, we have a lot of technical barriers - patents. Although Kyligence has many years of standardization accumulation in four industries: finance, manufacturing, retail, and medicine, the real barrier is not in technology, but in having China's largest group of digital transformation pioneer customers, who are also in leading positions in various industries and eager for innovation, so the speed and volume of commercialization are constantly growing.
Two major transformational directions for data analysts in the future
"Near Craftsman": AI is a great challenge for data analysts, many data analysts do their daily work is to make various reports, will they lose their jobs? What is the direction of the future transformation?
Han Qing: The transformation of data analysts should be very fast. Some will transform into prompt engineers, poring over metrics and better understanding the true business intent. Another transformation direction is governance, where enterprise data often has the problem of inconsistent caliber, which can help the business unify the caliber of indicators and optimize enterprise data governance problems.
"Near Craftsman": What kind of development talents will be needed in the future?
Han Qing: From the current perspective of AI development, the extension of talent changes has expanded. Previous development was more about writing programs, and more required professional ability, computer knowledge, and deep technical background. But in the future, prompt engineers do not need to be able to write programs, the most important thing is whether they can ask the right questions, ask the right questions, and the problem will be solved halfway. This community will be larger, and it may not be appropriate to call developers again.
"Near Craftsman" is an interview column launched by CSDN, which means "Approaching Craftsmen", approaching tool creators and technical managers who are deeply engaged in open source, cloud, AIoT, root technology, digital transformation, and cutting-edge technologies, understanding how they view the current development work, sharing the characteristics of their carefully crafted tools, and analyzing the development status and future trends of the entire industry.
To this end, based on open source, cloud, AIoT, root technology, digital transformation, cutting-edge technology and other fields, if you and your team have reporting needs, or if you have insights into technology trends, or new insights on in-depth application practices, scenario solutions, etc., welcome to contact CSDN to contribute, contact: WeChat (hanbb120, please note the contribution + name + company position), email ([email protected]).
Welcome to the "2023 AI Developer Ecosystem Questionnaire" launched by CSDN to share your real experience, and more beautiful gifts are waiting for you!