Baijiao is from the Au Fei Temple
Quantum Position | 公众号 QbitAI
This time, the large model can really allow humans to free their hands.
Today, the iFLYTEK Spark large model V3.5 is new in spring, directly poking at the pain points of office scenes!
- The Spark large model has been upgraded to support long text, long graphics and text, and long voice...... It can not only quickly learn massive texts, graphic materials, conference recordings, etc. from various sources, but also give professional and accurate answers in various industry scenarios.
- There is also an "intelligent twins platform" specially launched for enterprises to create intelligent assistants to solve the last mile problem of large-scale model application enterprises.
- In addition, iFLYTEK Xinghuo's voice interaction capabilities have been further upgraded, with the first multi-emotional super-anthropomorphic voice synthesis, AI can "emotional resonance", and functions such as "one-sentence voice reproduction" have also been launched.
Well, at present, the Spark model has been upgraded, so let's experience it for the first time.
Here comes the "Super Knowledge Assistant".
In the previous official trailer, everyone paid a lot of attention to the three "long" features that will be released. What are the considerations behind iFLYTEK's launch of this series of new features?
According to Liu Qingfeng, chairman of iFLYTEK, they have seen that for a period of time, developers and users of iFLYTEK Xinghuo have paid great attention to the acquisition and learning of knowledge.
In the process of knowledge acquisition and learning, the information that the majority of users can get is often not only ready-made long texts, but also the content of newspapers and magazines, the PPT content of various seminars, the board books on the teacher's blackboard, the notes of classmates, as well as various meeting recordings, interviews, various online press conferences, training and education videos, etc., can you upload these texts, pictures, voices, etc. to Xunfei Xinghuo to quickly obtain knowledge?
This requires large models to solve not only long text, but also long graphics and texts, long audio, and the accuracy of various enterprise and professional industry applications.
To this end, iFLYTEK has launched the first large model that supports long text, long graphics and long voices, to solve the needs of users for obtaining multi-source information in real scenarios.
Long text
According to reports, after the upgrade, the current Xinghuo large model's general long text capabilities, including long document information extraction, long document knowledge Q&A, long document summary, long document text generation, etc., have generally reached the level of 97% of the latest large model version of GPT-4 Turbo in April, and the overall level of long text of Xinghuo large model has surpassed GPT-4 Turbo in knowledge question and answer tasks in multiple vertical fields such as banking, insurance, automobiles, and electricity.
In order to cope with the problem of operational efficiency, the Xinghuo model has been specially pruned and distilled, and the 13B version has been launched, with an effect loss of less than 3%, but the efficiency in response time and generation effect has been improved.
Without further ado, let's get started with the actual review.
First of all, let's look at the first hors d'oeuvres, throw the first volume of Feynman's lecture notes on physics directly to him, and don't tell him what the book is called, just ask what the book says.
Hmm~ not bad, the general direction is right.
Long graphic text
Next, let's look at the second question to test its ability to recognize long images and texts.
According to Liu Qingfeng, the image and text recognition model has now covered 31 of the most common typical scenes, such as educational books, academic papers, patents, newspapers, posters, product white papers, PPT and menus, APP screenshots, speech photos, etc., as well as 18 layout elements (including headers, footers, titles, columns, paragraphs, tables, illustrations, etc.)
In this case, throw it a PPT of the "China AIGC Application Panorama Report" produced by the latest qubit think tank, and ask about the relevant details.
As a result, whether it is the market size, business model, investment and financing situation, they are all clearly answered.
Long voices
Finally, let's test its long voice ability. It can be seen that iFLYTEK Xinghuo can support a variety of audio and video formats, as long as it does not exceed 1GB in size.
Then throw it to him this time, iFLYTEK officially released a demonstration video of long voice ability.
Outcome:
Even "What did Liu Qingfeng do" was answered accurately.
Imagine that in the process of information acquisition or knowledge learning, the materials obtained are nothing more than ready-made texts, papers and books, or seminar PPTs, screenshots of notes, as well as various conference recordings, press conferences, online teaching videos, and so on.
The upgrade of the iFLYTEK Xinghuo model: the support of "long text, long picture and text, and long voice" can be said to cover the entire scene.
Using it, it is equivalent to everyone having a knowledge assistant, isn't this a proper little tool for learning and working~
Multi-emotional hyperanthropomorphic synthesis & one-sentence sound reproduction
In addition, there are also multi-emotional super-anthropomorphic synthesis functions, and the function of one-sentence reproduction is newly launched, which can be experienced directly on the Xinghuo APP.
At the launch of iFLYTEK Xinghuo V3.5 at the beginning of the year, iFLYTEK launched the super-anthropomorphic dialogue function, which has now been further upgraded, not only more realistic, but also richer emotional expressions, including happiness, sorry, comfort, coquettishness, confusion and other emotional expressions The perceptibility of emotional expressions has reached more than 85%.
At the same time, it has also launched a one-sentence voice replica, which can customize your AI assistant voice in one sentence. In this way, when you are on a business trip, you can also tell stories to your children, or read books and newspapers to your grandparents, bringing more warmth to the world.
Xinghuo agent platform
Office scenarios have long faced a pain point - how to efficiently acquire and learn knowledge. The intelligent twin platform launched this time is specifically for enterprise scenarios.
On the iFLYTEK Xinghuo intelligent twins platform, first of all, based on the Xinghuo large model, it will automatically realize the accurate understanding and task planning of user input. After analyzing the relevant tasks and corresponding tools, iFLYTEK Xinghuo has built a system of external information sources including weather, flights, and enterprise checks.
At the same time, through the mechanism of mutual authentication, the Xinghuo intelligent twin platform also realizes the opening of the OA system, CRM system and ERP system, which are often independent and isolated, and completes the corresponding operations.
In addition, the Xinghuo intelligent twins platform can also realize the creation of new agents and the collaboration of multiple agents by dragging and dropping. With the above set of combinations, we can quickly reach the last mile of the landing of large-scale model application enterprises.
The technical concept of iFLYTEK large model: starting from solving real problems
It can be seen that the upgrade of the Spark model is more grounded and solves the real needs, rather than just the upgrade of performance parameters.
On the one hand, in the actual experience of the iFLYTEK Xinghuo large model, it is a rigid demand scenario for enterprises.
According to Qimai data, the number of downloads of the iFLYTEK Xinghuo APP on Android has exceeded 96 million, ranking first among the domestic tool general model APP.
From the perspective of C-end usage scenarios, iFLYTEK Xinghuo's users are mainly concentrated in the office field. The peak usage period is concentrated at about 9:30 a.m. and 3 p.m. on weekdays, mainly in the Internet, scientific research, education, and media industries.
This logic of using technology to solve rigid needs is also reflected in the growth of many of iFLYTEK's businesses.
With the support of the large model, the annual revenue of the open platform and consumer business reached 6.19 billion yuan, becoming the largest business segment of iFLYTEK. The smart automobile, smart medical and smart finance business segments contributed revenue of 700 million yuan, 540 million yuan and 290 million yuan respectively, an increase of 52.2%, 14.9% and 26.1% year-on-year respectively. In the field of C-end intelligent hardware, the GMV of consumer hardware such as iFLYTEK smart office laptops, iFLYTEK smart voice recorders, and iFLYTEK intelligent translators equipped with iFLYTEK Xinghuo increased by 84% year-on-year.
On the other hand, this is also the technical concept that iFLYTEK has always conveyed to the outside world: while advanced technology continues to iterate, it has always been committed to solving realistic scenarios.
A typical example is that every time the large model is upgraded, iFLYTEK Xinghuo has new bright industry applications, such as the release of the first large voice model in January this year, and the debut of the image and text recognition large model. With the blessing of the base model, it constantly breaks through the boundaries of the ability of the large model.
However, each upgrade also corresponds to the application of actual scenarios, and truly solves real problems. For example, this time, Liu Qingfeng focused on the application in bidding, contract, education and other scenarios.
For example, in the bidding scenario, iFLYTEK and the National Energy Materials Corporation have cooperated in an intelligent unmanned evaluation system in the enterprise procurement scenario, which has been recommended as a typical case on the website of the State-owned Assets Supervision and Administration Commission.
The system will further superimpose long text and long graphic capabilities, which can make bid evaluation more convenient, efficient and accurate.
There are also contract assistants. It can conduct risk review, contract comparison, summary and contract generation of contracts, and quickly identify potential risks and vulnerabilities. In addition to the needs of work, it is also fully used in daily life such as buying and selling goods, decorating or buying insurance.
This large-scale model technology concept to solve real-world problems has also allowed iFLYTEK Xinghuo to quickly build a certain influence in the industry.
Since its release on January 30 this year, iFLYTEK Xinghuo V3.5, as the first large-scale model of nationwide computing power training, has been widely welcomed by partners and developers in various industries. Especially in some key industries and major strategic areas, the Xinghuo model empowers more and more industries, such as automobiles, home appliances, and operators, with the overall solution of "cloud, edge, and device...... Playing a role in the real economy.
From the perspective of developer ecology, in the past less than three months, iFLYTEK has added 550,000 real-name verified developers, more than half of which are from enterprises.
Large models support new quality productivity
This year is undoubtedly the first year of large-scale model application. The large model supports new quality productivity and helps enterprises in digital transformation.
But how should enterprises use it, how should they use it, and how should they use it? The development of the large model to the present, can roughly sort out these three models.
In short, the current large-scale model blessing of the general AI native APP, the function fragmentation can be limited to the tools that can be called each time, and it also depends on the model upgrade of each large model company.
There is also an open source model or access API, but there is still a long way to go between the general large model to land in the real application scenario, which requires the collaboration of technology and industry know-how, which is a big challenge for enterprises.
The second is the super APP, which integrates various AI native fragmentation capabilities to improve the efficiency of communication and execution in the workflow. However, if the internal data is not included to achieve the integration of internal and external knowledge, the efficiency improvement ability of the large model is limited.
And iFLYTEK Xinghuo showed the fourth mode this time - the agent platform. AI agents have become a definite trend as a means of improving efficiency for enterprises. However, iFLYTEK directly launched a productized solution, and the whole process has a low threshold, and can realize the construction of agents and the collaboration of multiple agents with a simple drag-and-drop, so that enterprises can more easily get started and use them directly, which helps to realize the large-scale implementation of agents and realize the inclusive value of large models.
Finally, it can be seen that more and more large model upgrades are going in a more grounded direction, which actually represents a specific trend.
That is, large models have made their way to our daily lives, and artificial intelligence is moving deeper and deeper in the direction of solving real-world problems.
— END —
量子位 QbitAI 头条号签
Follow us and be the first to know about cutting-edge science and technology