With just a few questions to answer, you can generate a personalized resume, provide optimization suggestions, and schedule an AI mock interview. Such an efficiency artifact appeared on the field of the second Search Innovation Competition.
The "Liver to the Morning" team from Nanjing University of Aeronautics and Astronautics won the championship of track 3 of this competition in one fell swoop with the AI resume assistant "Linchance Job" by using the Wenxin Yiyan plug-in platform "Spiritual Matrix" and the platform and technology provided by Search.
The Search Innovation Competition is a national AI technology competition jointly organized by Search, NVIDIA and NVIDIA, major universities, associations and scientific research organizations across the country.
Officially launched on September 7, the second Search Innovation Competition with the theme of "New Search, New Singularity" finally attracted more than 2,800 people to sign up after three months of competition, covering 95 double-first-class universities, and finally 28 teams won awards, 50,000 students directly participated in relevant course Xi, and the publicity covered 1 million developer groups.
Behind each outstanding achievement, it is undoubtedly inseparable from the dedication and efforts of outstanding talents, and the blessing and exploration of the industry. Xiao Yang, Vice President of the Group and Head of the Search Platform, said in his speech at the award ceremony, "In this wave of large models that are far from the end, the technological change gene of search must not only continue to play a role, but also need to evolve itself. And this evolution needs the support of innovative forces. Thanks to the platform of the Innovation Competition, the search technology has changed its genes and quickly collided with many young blood. ”
In this regard, He Tao, vice president of NVIDIA, also said in his speech at the award ceremony: "As a technology partner of the Search Innovation Competition, we are very pleased to see that the search competition has attracted more than 2,000 outstanding contestants. Over the course of the year, industries have undergone a platform transformation from general-purpose computing to accelerated computing and generative AI. He also encouraged all the students present to follow the trend of technology development, seize the opportunity, and become people who know how to use AI, so as to grasp the initiative in future career development. ”
The results of the five major tracks were impressive, and 28 teams stood on the podium.
Xiao Yang said that everyone can become a developer, and the arena of the Search Innovation Competition is the best venue for achieving excellence. This year's competition has set up five tracks, with topics related to search core business scenarios, such as semantic search, multi-modality, software and hardware combination optimization, and other rich technical directions, and set up a track three product track, calling on participants to "gain insight into the needs of users in search scenarios and solve them by building AI applications".
In the end, 28 teams stood on the podium.
In the first track of the "Search Answer Organization", the "none" team from the University of Chinese Academy of Sciences won the first prize. The team fine-tuned the LoRa fine-tuning LLM scheme, selected public Q&A data to enhance training, used large model distillation to improve the learning Xi effect, and used Neftune for noise embedding to increase the robustness of the model, so that the test data results were more in line with user needs. Each technology selection gives sound motivation, solid analysis, and credible conclusions. This reflects the team's in-depth understanding of the search for answers to organizational questions and excellent scientific research capabilities.
Track 2 "TOPK Search Based on Vector Intersection", "Dawn" team won the first prize. There is only one person in this team, and he is still a full-time dad, who is not only far ahead in the results of the machine evaluation, but also won the unanimous recognition of the judges in the innovation, versatility and advancement of the overall technical solution in the defense.
In track 3 "Designing an AI native application that solves the needs of search users", the "Liver to the Morning" team from Nanjing University of Aeronautics and Astronautics won the championship with the AI resume assistant "Linchance Job". With just a few questions to answer, you can generate a personalized resume, provide optimization suggestions, and schedule an AI mock interview.
Track 4 "GPU-based Approximate Nearest Neighbor Retrieval Algorithm Challenge" was won by the "HDU01" team from Hangzhou Dianzi University. Based on researching relevant algorithms in the industry and in-depth analysis of the open source version, the team made meaningful changes that made the final version more than twice the performance of the open source version.
In the fifth track of the "controllable generation algorithm challenge", the "lucky bottomless team" of Beijing Institute of Technology won the first prize. Through careful cleaning and enhancement of data, they provide a stable foundation for model training, and the multi-model fusion strategy they adopt reflects a good combination of theory and practice, which ultimately improves the relevance and quality of generation.
The competition also organized a total of 16 online and offline activities, providing more than 100 hours of knowledge training and communication opportunities, with more than 50,000 students directly participating, covering 100W developer groups.
The DNA of technological change is directly rooted in search.
It is the first major manufacturer in the world to launch generative AI, released the beta version of Wenxin Yiyan on March 16, and officially opened it to the whole society on August 31. At World 2023 held in October, many refactored products and applications were launched, among which the refactored search is particularly interesting.
The search reconstructed by the large model now has three new capabilities: extreme satisfaction, recommendation stimulation and multi-round interaction. When a user asks a question, the new search is no longer to provide a bunch of links, but through the understanding of the content, generate multi-modal answers such as text and dynamic charts, so that users can get satisfaction in one step, leading the generational change of the search industry. Currently, there are more than 300 million searches per day, which are met by generative search. The proportion of searches that give satisfactory answers at the top has nearly doubled.
Li Xipeng, general manager of NVIDIA's development and technology department in Asia Pacific, also elaborated on his understanding of "new search", "From the earliest book retrieval, data retrieval, to the retrieval of digitized content, to the current LLM-based generative retrieval, it shows that a good field can continue to lead innovation, continue to generate innovation, and continue to absorb innovation." The search business itself is a collection of cutting-edge deep learning such as NLP, CV, and voice. Based on this, the large model has innovation, and the first landing scenario is search. It is very worth looking forward to the large model shining in the search engine. ”
The search for technological change genes is injecting young blood.
In his speech, Xiao Yang said, "In the face of the wave of large models that are far from the end, the search for technological change genes must not only continue to play a role, but also need to evolve themselves. And this evolution needs the support of innovative forces. "The power of innovation refers to talents, and the search innovation competition builds a stage for selecting talents, as the industry says, to create the "Olympics of the search industry".
With the platform of the Search Innovation Competition, the technological change gene of search is injecting young blood. Xiao Yang said, "In the coming period, search will continue to embrace young inspiration and inject new vitality into its technological genes." ”
NVIDIA's developer community is also a place where the power of innovation comes together. He Tao said, "NVIDIA also provides developers with a lot of resources for learning and Xi exchanges, NVIDIA developer community, there are currently 4.5 million developers around the world, CUDA acceleration library ** has more than 48 million times, we very much welcome students and contestants to join the NVIDIA developer community, enhance self-competitiveness, accelerate your career success." ”
Build a platform that aggregates ideas, scenes, and technologies, attract more fresh inspiration, and the emergence of technologies that can be implemented.