Jin Lei from Au Fei Temple
量子位 | 公众号 QbitAI
"Report to the boss! These employees are lazy, and they rely on large models for their work. ”
However, in the face of seemingly opportunistic behavior, the boss was unusual, and nodded again and again with great satisfaction:
I'll give you 5 minutes, can you write a popular science article based on a popular topic?
They can.
I saw these employees roll up their sleeves and start to practice.
Let's start with a hot topic – autonomous driving: a new chapter in the future of mobility.
Later, when you go to the creation interface, you can see that the "topic description" has been automatically generated, and the copywriting type is selected as "knowledge popularization".
Finally, just click on "Generate Article" and you're done.
And it's the kind that only needs to wait for about 20 seconds to write three articles in one sitting!
Not only that, but even expansion, illustration, typesetting, and publishing have become "dot-by-dot" things.
So what is the origin of these employees, who dare to use large models to work so clearly?
Don't sell off.
They are the Xinghuo office digital employees just released by iFLYTEK, and behind their capabilities are the blessing of the iFLYTEK Xinghuo model.
And unlike the traditional automation tools in the past, this time, iFLYTEK Xinghuo integrates the AI Agent capabilities of Dahuo.
As a result, it is not only possible to generate articles and pictures, but also to complete tedious tasks such as data analysis and intelligent operation.
Even make a 40-page year-end summary PPT, just wait a few minutes.
△ Original speed playback
As for what exactly Spark Digital employees want to do?
Spark Digital Staff · Every office aspect has a new look
However, office is only a corner of the scene that Spark Digital employees are involved in.
Digital employee onboarding to assist with management and marketing
If the digital employees of Xinghuo Office belong to the players of process-oriented work, then let's take a look at the digital employees of Xinghuo Management to assist various audit and decision-making scenarios in the operation and management process of enterprises.
Specifically, three products were released under the Spark Management Digital Employee section, involving audit-oriented digital employees and decision-making digital employees.
In the audit scenario, there are digital employees for intelligent bid evaluation and digital employees for smart contracts.
For example, in the process of bidding and procurement management, it involves long procurement business processes, many information systems, procurement project members, bid evaluation members face problems such as tight time and heavy tasks, such as the preparation of bidding documents need to be repeatedly revised and reviewed, and the bidding documents need to be checked and evaluated item by item in the bid evaluation process. The whole process is time-consuming and labor-intensive.
With the blessing of the intelligent bid evaluation digital staff, the key links can become intelligent and automated, and the bidding document evaluation method and the response content of the bidding documents can be automatically identified through the Xinghuo model, and the automatic comparison and judgment is carried out; the quantitative and qualitative bidding content is fully automatically reviewed, and the subjective bidding content is simulated reviewed, and the response content to be scored is simulated and scored; the bid evaluation experts review and confirm the evaluation and scoring results based on the evaluation and scoring results of the digital employees. Establish a two-dimensional evaluation system of AI + experts to make bid evaluation more accurate, efficient and convenient.
And from these processes, it is not difficult for us to find that the pursuit of intelligent bid evaluation digital employees is not only to automate work, but to "assist" human employees, plus human knowledge and experience, to achieve the effect of efficiency and professional double buff superposition.
△ Human-machine collaboration to jointly enhance professionalism
Similarly, there are smart contract digital employees.
It covers the whole process of contract drafting, review and approval, signing, performance and archiving, and there are a large number of templates that can be used out of the box, and can be retrieved by dialogue.
△
Dialogue for communication
The same is to automate and intelligentize contract-related work, so that efficiency goes straight to Pro Max.
In the enterprise management decision-making scenario, iFLYTEK also released the decision-making digital employee - generative smart cockpit.
Overall, this is an intelligent interactive system that integrates big data, large models and virtual humans.
It is like a large market, which can include the development status of the company's related business in one map, not only with real-time data display, but also with digital employees answering questions online and gaining insight into the subtle changes in the business.
Moreover, the generative smart cockpit can already cover multiple industries, including finance, cultural tourism, energy, FMCG, manufacturing, transportation, and media.
However, the ability of Xinghuo Digital employees is not only to "play assistance" to improve efficiency, but also to become mentors and train employees.
Marketing digital employees, for example, are playing such a role.
He can extract a number of keywords from a sales employee's conversation with a customer, and score and analyze his words in the marketing process.
Deficiencies can also give suggestions for improvement to help employees improve their own capabilities.
This is the strength of Xinghuo Digital employees.
Then let's return to the essence of the problem, the reason why iFLYTEK has landed a series of digital employees is essentially to solve the digital transformation problems faced by enterprises in the context of economic prosperity. So let's take a look:
How Spark Digital Employees Reinvent the Way of Digital Transformation
Throughout the conference, whether it is process-oriented, audit-oriented or decision-making Spark Digital employees, they are all in line with and based on iFLYTEK's own set of ways of helping digital transformation.
From the perspective of the lowest base, Xinghuo digital employees rely on the multimodal capabilities of the Xinghuo large model.
It includes text generation, language comprehension, quizzes, logical reasoning, math skills, and coding skills.
The data to be processed by these technologies is the data assets from the enterprise, such as production data, business data, etc.
On top of the technical base, there is the Xinghuo digital employee platform, which includes four centers, namely the process planning center, the user interaction center, the data analysis center and the domain model center, which can handle different tasks accordingly.
Finally, the product is formed in a simple and easy-to-understand interactive form, which is the case we just showed.
It is worth mentioning that in this whole set of play, what iFLYTEK wants to do is not to completely change the enterprise, but to empower.
This enabling breakdown can be divided into three parties:
Enterprises: On the basis of making full use of old information, they need to continue to innovate with the help of new technologies to reduce costs and increase efficiency;
Customers: It is necessary to play with digital and intelligent marketing methods to reach target customers more accurately;
Employees: Need to use the power of AIGC and derivatives to help them improve their work efficiency.
This is the way of iFLYTEK's digital transformation, so under such a new paradigm, it also raises a new question for the public:
Digital employees, is it really reliable?
Unreliable, look at the data.
In fact, today's announcement of iFLYTEK's announcement of Xinghuo Digital employees is not the first time, but as early as May 2023, it will be released in a conceptual way.
Therefore, the data disclosed after more than half a year, whether it is the digital employees of Xinghuo Office can reduce the creation from "days" to "minutes", or the digital employees of Xinghuo Marketing can improve the efficiency of quality inspection by 22 times in the field of automobile retail, etc., it can be said that they have passed the time test.
In addition, there are a lot of data extracted from "real gold refining" in real scenarios like this, taking the management of digital employees as an example:
- 47.3% improved the operation and management ability of enterprises;
- 48.1% improved supply chain management capabilities;
- Reduce enterprise labor costs by 23.8%;
- 32.5% access to low-cost procurement channels;
- Get more market orders42.3%.
In the more than half a year from the release of the concept to the proof of concept, ChatGPT-like and the large model behind it are in a cooling-off period: every company is thinking about how to make them truly energy efficient.
After all, the difficulty of "getting on the job" of large models in industrial applications is more challenging, which not only requires the professionalism of domain knowledge, but also the ease of operation, how much efficiency improvement it can bring, etc., are all included in the considerations.
iFLYTEK has used more than half a year to hand over this digital employee answer sheet with "real guns", which proves that the industry can play a role in reducing costs and increasing efficiency with the blessing of large models.
After looking at the data, the second look at the technology.
Since ChatGPT detonated the 100-model war, iFLYTEK Xinghuo has always occupied a place in the first echelon of domestic large models.
The reason why it can be done at the time of the boom is inseparable from the technical "fit" between iFLYTEK and the large model.
The strength of large model products lies in the ability to understand language, question answer, logical reasoning and other capabilities at the same time, which are necessary to achieve cognitive intelligence.
iFLYTEK has been deeply engaged in these key technical capabilities, including natural language processing, audio and video processing and other multimodal capabilities.
Therefore, in the evaluation lists of large model performance at home and abroad, iFLYTEK Xinghuo can also be seen.
For example, in the third annual evaluation report on large models (Artificial Intelligence Large Model Experience Report 3.0) released by the China Enterprise Development Research Center of the Xinhua News Agency Research Institute, iFLYTEK Xinghuo won the championship with outstanding results, and won the first place in the three evaluation indicators of basic ability index, IQ index and tool efficiency improvement index.
The results of the list are important, after all, it is one of the indicators to test the performance of large models, but for iFLYTEK, it seems that the more important thing is "how well it is used".
From this point of view, it reflects that in addition to iFLYTEK's "first-mover advantage" in technical books, it is also the same in the landing of scenarios.
Previously, iFLYTEK has accumulated profound scenarios in AI applications in China, such as education, medical care, finance, automobiles, services and other fields, and has built deep industry barriers.
After all, in the era of large models, the value of data has never received so much attention as today.
Grasping technology with the left hand and grasping the ground with the right hand, based on this logic, you can also get a glimpse of the development of the iFLYTEK large model.
But what is more worth mentioning is the speed of iFLYTEK Xinghuo under this logic: fast, really fast.
In February last year, iFLYTEK responded to the creation of ChatGPT for the first time and gave a clear time point. Since then, various major actions have also been in full swing:
- In May, iFLYTEK Xinghuo V1.0 was officially released, and directly demonstrated the five major application achievements;
- In June, Xinghuo was upgraded to V1.5;
- In August, iFLYTEK Xinghuo V2.0 was released, and multi-modal capabilities were realized.
- In October, iFLYTEK Xinghuo V3.0 was released, and at the same time, 12 major industry models were released, and the general artificial intelligence industry ecology began to take shape, and it was announced that it had fully benchmarked ChatGPT and surpassed GPT-4 in medical care.
Technology is iterating, landing is promoting, and behind such speed is the two-way parallel model of iFLYTEK.
After data and technology, the last thing that needs to be looked at is the trend.
In the qubit "Virtual Digital Human Deep Industry Report", we have given a prediction:
In 2030, the overall market size of virtual digital humans in mainland China will reach 270 billion.
Among them, there are about 175 billion identity-based virtual digital humans, and the total scale of service-based virtual digital humans exceeds 95 billion. At present, the market is still in the early stage of cultivation.
Not only that, applications such as digital employees can use their powerful AI technology to reduce costs and increase efficiency in processes, decision-making, and services for enterprises, which is already a critical part of the digital transformation wave.
Under such a market scale and trend, digital employees have long become a must for major players in China alone, including Baidu, Alibaba, Tencent, Xiaoice and so on.
Of course, iFLYTEK has already made efforts on digital humans and has had a series of release actions, and this time iFLYTEK Xinghuo is integrated into it, and whether it is positioning or subdividing the track more accurately, what is the meaning of this?
First of all, in terms of technology development trends, large models have already ushered in a new development trend at the end of the year and the beginning of the year, one is the transition from large language models (LLMs) to multimodal large models (LMMs), and the other is the rise of AI agents on this basis.
The two can be said to bring new life to applications such as digital humans and digital employees, and bring capabilities and results to a higher level.
Secondly, for such a layout and planning, iFLYTEK also said:
In the past, one of the major difficulties in the deployment of digital employees was the migration problem in different fields, but now, with the advent of large models, the "playing style" has become universal, which greatly reduces the cost of migration.
This means that the widespread promotion of digital employees and the acceleration of enterprise digital transformation have ushered in an excellent technical opportunity.
Finally, it is not difficult to see from the trajectory of iFLYTEK's large-scale model development over the past year or so that it seems that all the previous flags and released concepts have been gradually realized.
Based on this, the reality of the sentence "fully benchmarking GPT-4 in the first half of 2024", and what surprises will be brought by digital employee capabilities with the blessing of new capabilities, are worth looking forward to.
One More Thing
At the scene, iFLYTEK also showed a more interesting way to play.
In the "AI Personality" app, start a conversation with Cao Cao, and he will answer questions about "digital transformation" in Cao Cao's tone:
Well, it's kind of interesting.
— END —
QbitAI · Headline number signed
Follow us and be the first to know about cutting-edge technology trends