Author: Hao Junhui Source: IT Times
Can large models liberate workers? Tencent tries to answer this question.
On April 22, Tencent announced that its collaborative SaaS products have been fully connected to Tencent's hybrid model, in addition to "one door and three masters" products such as enterprise WeChat, Tencent Meeting, and Tencent Docs, Tencent Lexiang, Tencent e-Signature, Tencent Questionnaire, Tencent Cloud AI code assistant and other collaborative SaaS products have been connected to Mixyuan.
As the last Internet giant in BAT to release a general model, Tencent's hybrid model has expanded to a trillion-level parameter scale, but it is relatively cautious in external promotion, and has always focused on polishing product capabilities and promoting implementation practices.
"Since its launch, Tencent Hybrid has highlighted its practical label, and one of its core missions is to optimize the user experience of Tencent's products. At the press conference, Tencent announced that external individual developers and enterprise customers can directly call Tencent's hybrid capabilities through Tencent Cloud's APIs, which can be combined with SaaS products to solve users' pain points in specific scenarios.
Obviously, Tencent's large model has chosen the route of "from the inside out" and "from B to C", and the internal incubation is mature and then enters the commercial field. Up to now, more than 400 businesses and scenarios within Tencent have been connected to Tencent Mix, and five collaborative SaaS products were shared at this communication meeting. Zhang Feng, the person in charge of the application of Tencent's hybrid model revealed that in the internal test, the API call volume of C-end products is very large, and the access effect is also good, and the hybrid element is ready to face more than one billion real users externally.
Tencent Lexiang AI Assistant: Increase the efficiency of knowledge transfer for frontline employees
Tencent Lexiang, Tencent e-Sign, Tencent Cloud AI Code Assistant, Tencent Meeting, Tencent Docs and other Tencent's collaborative SaaS products are several online applications commonly used by "migrant workers".
AI assistant is currently a common method for large models to be applied, and the key to success lies in whether there is high-quality data and knowledge base.
Tencent Lexiang is an internal knowledge learning and collaboration platform that provides enterprises with learning and training, knowledge management and culture building services, and has served more than 300,000 customers and more than 100 million authorized users.
At present, the Lexiang AI assistant has high practical value in two types of scenarios. Kuangzhen Law Firm in Hunan Province has imported more than 40,000 internal knowledge documents compiled by employees into Tencent Lexiang to build an exclusive AI knowledge base and realize the AI intelligent Q&A function, and employee surveys show that AI satisfaction with typical questions is as high as 93 points, and the end-to-end question accuracy rate is 91%.
For example, there are more than 2,000 grid members in Nanshan District, Shenzhen, who need to master a large number of complex rules and regulations and work processes when dealing with various disputes in the region.
Zhou Zhizhi, general manager of Tencent Lexiang, introduced that at present, when Lexiang accesses the mixed element, only the intelligent question function is fine-tuned based on the mixed element, and many other functions, such as intelligent creation, intelligent refinement of course knowledge points, intelligent writing activity plan, etc., are all called to the API capabilities of the mixed element.
AI Code Assistant: Can You Replace a Programmer?
The tornado of large models set off by OpenAI quickly swept into these Internet giants.
Tencent Cloud AI Code Assistant is a digital tool for intelligent completion and generation of positioning codes, which is based on Tencent's professional coding model, which can provide high-quality code completion based on multi-file understanding and context. Just like there is a master next to him, when you encounter technical problems, you can ask directly, and the code is out of the way? The AI code assistant can not only find out the problem, but also give suggestions for modification, and the quality of the code will rise.
It is understood that Tencent Cloud's AI code assistant has achieved more than 40% coverage of development employees within a cooperative enterprise, with a code generation rate of about 30%, an adoption rate of about 20%, a storage rate of more than 60%, and a completion time of less than 500ms. Combined with Tencent Cloud's internal experience in large-scale production, the coding efficiency has been improved by more than 20%.
Tencent Cloud AI Code Assistant can provide developers, development teams, and enterprise customers with AIGC services for coding scenarios with security, compliance, and high-viscosity experience, with functions such as code completion, technical dialogue, code diagnosis, and unit testing.
Tencent Docs: "One sentence" directly helps write PPT
When OpenAI just released ChatGPT, a migrant worker jokingly said: Artificial intelligence that can't do PPT is useless. Nowadays, users can use Tencent's Docs intelligent assistant to generate, beautify PPT and even generate notes during speeches.
Xiong Fei, product director of Tencent Docs, said that Tencent Docs is also exploring optimized application scenarios, such as users directly sending an Excel to the AI assistant, hoping to ask questions, analyze and answer based on complex data, so as to guide the next decision.
With the gradual digital transformation of domestic enterprises, in addition to traditional documents, professional documents, local document editing, papers, government documents and other documents that require special formats, the application demand for collaborative documents is also gradually increasing.
Commercialization of large models
SaaS has always been in an awkward position in China, and some people have even raised the question of "whether China wants SaaS". The reason for this is that it has been difficult for domestic enterprises to break through their willingness to pay. After the emergence of ChatGPT, it is believed that it will open the bottleneck between SaaS and demand, and help enterprises build an AI pass.
However, from the current point of view, various companies, including Microsoft, have not given a completely clear answer sheet for reference regarding the commercial value of AI applications. "The ability window period and commercialization logic of large-scale models to meet the commercialization standards also require verification time from the market and users. Xiong Fei, product director of Tencent Docs, said that the effectiveness and value of AI must be based on the optimal effect of users and scenarios, rather than AI for AI's sake.
It is understood that at present, several SaaS collaborative products with access to mixed elements have adopted different charging models. AI Code Assistant is a plug-in product, priced and sold separately, and the pricing of Lexiang AI Assistant will also be based on the original version of Lexiang to add value, such as a certain percentage increase based on the original version, providing a solution for additional purchases.