Editorial Department Organized from meet2024 qubit | qbitai
In the past year, the large model has developed from the technical competition of the "100 model war" to the competition of commercial value.
At the Meet 2024 Intelligent Future Conference, Liang Zhihui, vice president of 360 Group and head of 360 large model application, shared his experience and views on the application practice of large models this year, he said:
When promoting the application of large models, it is necessary to give full play to the advantages of large models, such as content generation, avoid its shortcomings, such as insufficient long-term memory ability, and realize human-machine collaboration. Rational use of the advantages of the large model to make it an intelligent assistant for employees can greatly improve the work efficiency and output of the enterprise.
In order to fully reflect Liang Zhihui's thinking on the application scenarios of large models in enterprise production, Qubit edited the content of his speech on the basis of not changing the original meaning.
About Meet Smart Future Conference: Meet Conference is a top business summit in the field of intelligent technology hosted by qubits, dedicated to the implementation and industry application of cutting-edge technology technology. This year, a total of dozens of mainstream ** and live broadcast platforms reported and broadcast the meet2024 conference live, attracting more than 3 million industry users to participate online, and the total number of ** on the whole network has exceeded 20 million.
Presentation Takeaways. In the era of large models, human listening, speaking, reading, and writing abilities can be enhanced, and large models and human beings are a relationship of enhancement rather than replacement, and they are truly proficient in prompt engineering to read 100 articles and write 100 prompt words. Intelligent customer service (the following is the full text of Liang Zhihui's speech).
Generative AI is not a panacea, and new ways of human-machine collaboration are needed.
Today, I would like to share with you some experiences and cases of 360 in the application scenarios of large models and enterprise production.
First of all, in our view, the relationship between AI and people in the era of large models is not to replace, but to be enhanced.
Whether it is for internal office or external marketing, large models can greatly improve people's reading speed, writing speed, and information search speed.
But we also want to remind you that generative AI or generative large models are not a panacea.
First of all, the big model still has the problem of hallucination, and it may make up some facts. Second, many large models lack industry knowledge in many scenarios and cannot accurately answer very professional questions. Third, prompt engineering can also cause confusion to users of large models. According to our data, if a person wants to be truly proficient in prompt engineering, he or she needs to read 100 articles** and write 100 prompt words before he can truly learn.
Some time ago, OpenAI launched GPTS, and it can be seen that the prompt word template behind it is very complex, and it is impossible to make every employee of the enterprise become an AI enthusiast before the large model can be promoted.
In this scenario, we hope to have a new way of human-machine collaboration.
From AI "elementary school students" to autonomous agents.
It is difficult to generate high-quality content from large models, and if you want to promote large models in your work, you must make use of your strengths and avoid weaknesses, and avoid things that large models are not good at.
We think the most appropriate way is for the big model to become our assistant, to become everyone's assistant.
The strengths of large models are content generation and content understanding, and the disadvantages are not only hallucinations and long-term memory, but also weak task planning ability, which is not as good as a human.
Suppose a college student is recruited as an assistant, training is required for entry, and the completion of actual work also depends on the workflow summarized by predecessors. We want to make it possible for large models to execute workflows in a step-by-step manner, breaking down tasks to get the job done.
Since the domestic 100-model war has opened up the situation of AI native applications, we have seen a lot of robots that ask and answer questions, and such robots are like primary school students.
Through the prompt word "hypnosis" model, for example, play a marketing director, and answer according to the marketing director's routine. But it doesn't understand your product, it doesn't know your company, and it doesn't understand how it collaborates, and the generated company introduction copy is basically unrealistic and unreliable.
The intermediate state is to combine the large model and the industry knowledge base to make it a knowledge assistant like a college student, but in this scenario, there is still only general knowledge, and the business of the specific company is still not understood.
We want the large model to become an autonomous agent, that is, an agent, that has multiple skills, industry knowledge, and can use multiple tools.
This agent has the knowledge background of the entire Internet to do it, and can skillfully use a variety of tools, which can help you check air tickets, book air tickets, know today's weather, and even know today's exchange rate and other real-time information.
A one-question-and-answer chatbot can only answer some simple, static questions, and cannot be integrated.
If you combine the large model with the agent, it will add a lot of skills compared to the previous two application forms, which can help you search, read, and write. The search can help you think of more keywords and complete the problem. After finding more information, the agent can read hundreds of web pages and finally complete a work report.
Three major scenarios for large-scale commercial applications.
Based on the large model and agent, we currently focus on three major scenarios: intelligent marketing, intelligent office, and intelligent customer service.
For intelligent marketing, we have both Wensheng Wenruxin** writing, Wensheng pictures such as e-commerce pictures, as well as Wensheng short** and Wensheng live broadcast.
In terms of Wensheng text, many large models say that they can do Xiaohongshu copywriting, but the more people who want traffic, the more they will combine Xiaohongshu copywriting with hot spots, in this scenario. Building alone may not be enough.
The Xiaohongshu copywriting tool provided by 360 will know what keywords are popular in Xiaohongshu now, can combine hot words with content, and call search capabilities to understand the key details of the product, and the quality will far exceed the copywriting generated by ordinary large models.
Based on the large model to generate different styles of copywriting, you can also generate countless different versions, the content repetition rate will be greatly reduced, and the chance of being recommended by the platform will be greatly improved.
At the same time, we see that many people are doing the promotion of **content, but whether it is the generation of **scripts or the generation of ** itself, the cost is very high.
In this scenario, 360 has a digital human generation scheme, and everyone can generate their own digital human to make. For example, in the cultural tourism scene, we can have a number Zhuge Liang to introduce the local situation in Xiangyang.
Secondly, we will see that many CEOs of companies are very busy, going to various places to give speeches and even attend cross-border events. In this scenario, not everyone can master multiple Chinese, so our digital human can allow CEOs to have face-to-face conversations directly with people, and even speak English.
With these capabilities, we can also let the digital human host the entire event, without worrying about the digital human saying the wrong word. can also be an anchor, use digital humans to open multiple live broadcast rooms, different live broadcast rooms can say different things, respond to the feedback of different audiences, and are no longer repeaters.
In terms of smart office, 360 also has several examples to share.
In terms of office, the highest praise for people is ten lines at a glance, and people can write a 10,000-word long essay at most a day. But with the big model, these everyday office writings are no longer a problem. In the same company, the efficiency difference between large-scale model-assisted writing and manual comparison is very large.
At present, 360 provides pre-programmed tools for many office scenarios, in which you can generate a PPT outline in 30 seconds, let AI help you complete SWOT analysis with one click, or spend 3 minutes to complete a 30-minute speech, everything in the past required ten or more than 100 times the time to complete.
Based on the large model, you can actively do search, you can also read the content you provide, and you can use your historical model as training data, so that the content written by the large model is more connotative, logical and factual.
There are endless reports and industry content to read every day, and in this scenario, you can let the large model help you read the entire report, summarize the report, and write the report.
For intelligent customer service, the biggest difference between a digital person based on a large model and an ordinary digital human is that it sells an interactive digital human with natural dialogue ability, which can not only be customer service, but also a clone of sales and even bosses.
In scenic spots, government affairs and financial scenes, you can let the digital person based on the large model to receive guests, and even become a sparring officer, as long as the enterprise puts the relevant information, whether it is audio or text content, after uploading, the large model can quickly understand the content, and finally become a virtual tour guide and corporate customer service.
Digital people can appear on a variety of large screens, all-in-one machines, APP, H5, and even let it make product recommendations, before 360 cooperated with a technology industry customer to recommend a product recommendation officer digital person, for example, you can recommend a laptop suitable for students.
That's all for today's speech, if you are interested in 360's products, our AI digital employees and digital humans have trial versions, welcome everyone to experience, thank you.