Author: Hao Junhui Source: IT Times
This is an interview that requires speeding up speech and increasing decibels to "grab" questions.
On the afternoon of July 7, before the 2023 World Artificial Intelligence Conference and Tencent Forum, Wu Yunsheng, vice president of Tencent Cloud, head of Tencent Cloud Intelligence, and head of Youtu Lab, accepted a group interview with the media in a small and noisy conference room. Nearly 20 days ago, Tencent officially released the MaaS panorama, cutting into the hot "big model track" with the industry's large model, and reporters were eager to know why Tencent avoided the general model and chose a path that looked more "realistic".
"What companies need is to actually solve a problem in the actual scenario, not solve 70-80% of the problem in 100 scenarios." Wu Yunsheng said that from the perspective of the company's strategy, Tencent is more focused on solving practical problems, while the general large model cannot completely solve all the problems of users.
Tencent, which has the largest number of individual users in China, has put the AI-based industrial Internet as the first step in its change when a new wave of artificial intelligence comes.
Tencent Cloud MaaS was upgraded again
On June 19, Tencent Cloud announced for the first time the R&D progress of Tencent Cloud industry large models, relying on the Tencent Cloud TI platform to build a selected store for large models in the industry, providing customers with one-stop MaaS (Model-as-a-Service) services, and has provided more than 50 large model industry solutions for more than 10 industries such as media, cultural tourism, government affairs, and finance.
At this World Artificial Intelligence Conference, Tencent Cloud once again announced a number of upgrades.
Among them, the newly upgraded Tencent Cloud's self-developed Xingmai high-performance computing network can improve GPU utilization by 40%, save 30%~60% of model training costs, and bring 10 times the communication performance improvement to AI large models. Based on Tencent Cloud's next-generation computing power cluster HCC, it can support a large computing scale of 100,000 cards. Tencent Cloud AI native vector database supports up to 1 billion vector retrieval scale, with latency controlled in milliseconds, which is 10 times higher than that of traditional stand-alone plug-in databases, and has a peak capability of million-level queries per second (QPS).
In terms of application innovation, Tencent Cloud's industry large model capabilities have been applied to scenarios such as financial risk control, interactive translation, and digital sapiens customer service, greatly improving the efficiency of intelligent applications.
The financial risk control solution supported by the industry's large model has improved efficiency by 10 times compared with the previous one, and through Tencent's accumulation of more than 20 years of black and gray industry confrontation experience and thousands of real business scenarios, the overall anti-fraud effect has been improved by about 20% compared with the traditional model. In the field of digital sapiens, this year Tencent Cloud launched a small-sample digital human factory, which can replicate 2D digital twins within 24 hours with only a small amount of data, greatly reducing the cost of enterprise application digital sapiens services.
"In fact, in the past half a year, we have been thinking and exploring, what is the most essential logic behind the combination of big models and various industries? In fact, there are only two points: first, the fundamental starting point of technology is to solve practical problems, and second, if you cannot explore the industry deeply, you cannot really solve the problems faced by the industry. The "test" brought by the real scene to the big model made Wu Yunsheng feel a lot.
Intelligent customer service is recognized as the most applicable industry for LLM (Large-Scale Language Modeling). At this conference, Tencent created an industry-large model for an online travel OTA company, and the customer-specific model after fine-tuned can solve business problems end-to-end without configuring the dialogue process. Improve task completion rate and reduce conversation construction costs. But in fact, getting a big model to really understand the customer's problem is not as simple as it seems.
"In the process of communication, the customer's thinking is jumpy and changing. For example, he just proposed to book the hotel on the 10th, but before the machine could answer, he suddenly said, help me see the hotel and flight on the 11th, and when the AI is still responding to the second demand, he may say, show me the twin room. Wu Yunsheng pointed out that it is still quite difficult for large models to achieve multi-intent recognition, and general large models cannot be solved simply, but need to combine scenarios, especially to reconstruct some very complex models when interacting with customers' systems.
The era of "dancing with group models" has arrived
After the initial hustle and bustle, how to commercialize AI large models, how enterprise customers can enjoy this round of AI dividends, and solve "AI anxiety" has become a hot topic at this World Artificial Intelligence Conference.
Zheng Qingsheng, partner of Sequoia Capital China, has entered the investment field since the middle of the PC Internet, and in his view, the winners of each era are derived from the technological natives of that era, such as the PC Internet era, where people value e-commerce, and social software has become the biggest winner; Since the beginning of the mobile Internet era, people have favored social software and long videos, but short videos occupy the most time, "Now we don't know which primary scenarios generated by AI itself will change our basic behavior." ”
Although it is unknown when the "killer app" of AI native will appear, "entering the game" must be the first step. Among the more than 30 large models unveiled by WAIC this time, in addition to the first round of general large models such as Baidu Wenxin Yiyan, Ali Tongyi, iFLYTEK Xinghuo, and SenseTime, the latecomers basically focused on the industry large models.
"For customers, the enterprise-specific large model with few parameters, low investment and quick results is easier to accept, and the willingness to pay is relatively clear." A startup exhibitor told the "IT Times" reporter that some bank customers who have used large models to transform customer service systems usually choose a private domain deployment method that integrates software and hardware, using their existing knowledge graph and data to train and implement reasoning, which not only ensures data security, but also the cost of investing computing power will not be too high. ”
"The industrial scene has become the best training ground," Tang Daosheng, senior executive vice president of Tencent Group and CEO of the Cloud and Smart Industry Business Group, said at the WAIC plenary meeting - industry development forum, choosing a cloud vendor with one-stop industry large model service capabilities to cooperate and build their own exclusive model based on the industry large model may be a feasible path for enterprises to explore the application practice of large models.
This means that the future will be an era of "10,000 models coexisting", each enterprise will have its own big model, and Tencent has decided to be the enabler of the new era.
In the MaaS service panorama released by Tencent Cloud last month, it was proposed that based on the Tencent Cloud TI platform to build a large-scale industry model select store, Tencent Cloud can provide more than 50 solutions in 10 major industries such as finance, cultural tourism, government affairs, medical care, media, and education. At the same time, Tencent Cloud has launched the Industry Large Model Tuning Solution to help model developers and algorithm engineers solve model call, data and label management, model tuning, evaluation, testing, and deployment in one stop, reducing the pressure of creating large models.
Based on these models and tool platforms, enterprises can quickly generate their own "exclusive models" by adding their own scene data.
"It's still the early stage of big model development, and I personally hope that a hundred flowers will bloom and everyone will try different possibilities in different fields." Wu Yunsheng believes that the development of artificial intelligence is a huge data project, which requires both common knowledge and professional and profound authoritative knowledge organization, and requires the joint efforts of all parties to truly make technology serve the industry.
AI for Science captures cosmic "flickering"
Of course, in addition to playing an effect on the digital transformation of industries, Tencent Cloud's big model industry model has also accelerated the application of AI technologies such as big models in the field of scientific computing.
Since 2021, Tencent, the National Astronomical Observatory, and the School of Computer Science and Technology of Fudan University have jointly launched the "Star Exploration Program", using cloud + AI to help China Sky Eye FAST process the huge amount of data received every day, and find fast radio bursts and pulsar clues through visual AI analysis, and 30 pulsars have been discovered so far.
At WAIC this year, Tencent announced that the star exploration program has made progress again, discovering two fast radio bursts for the first time through AI technology.
Fast radio bursts are a mysterious astronomical phenomenon that radiates energy released by the sun throughout the year every 1 millisecond, "flickering" the universe. But it "flashes" very frequently, the time is extremely short, and it is very easy to ignore and capture in the massive data, and it was not until 2007 that humans discovered the first one, 40 years later than the discovery of pulsars.
Compared with pulsar exploration, finding fast radio bursts at lower frequencies in massive data requires AI models with higher accuracy and faster calculation speed. In order to improve the computing speed, Tencent has specially designed a new, end-to-end AI algorithm for the exploration of fast radio bursts. Under the same computing power, this new astronomical data processing paradigm promotes signal processing efficiency to be 1800 times faster than conventional processing processes.
Previously, before AI reading the map, it was necessary to complete complex astrophysical preprocessing of the signal map, such as Fourier transform, dispersion... These jobs are specialized and complex. Now Tencent Youtu has created an "end-to-end AI algorithm" for astronomical data processing, which can skip the pre-processing step and directly enter AI recognition, greatly improving efficiency.
FAST generates hundreds of terabytes of data every day and tens of millions of signal maps per week. In the face of massive data, through the "multi-example learning method + attention mechanism", Tencent Cloud can quickly locate and identify useful information in the data, and provide strong underlying computing power support.
Today, Tencent Cloud and FAST are continuing to detect radio signals in the constellation Andromeda, 2.5 million light-years away, and expect more "cosmic scintillation" to be captured in the near future.