SenseTime – Officially Launched the Large AI Model SenseNova
On April 11, SenseTime Group Inc. (Chinese: 商汤科技) held a technology exchange day event, "SenseTime TECH DAY," during which it announced the launch of SenseNova (Chinese: 日日新), its suite of new large AI models. SenseNova is developed with the company's access to vast data and deep computing power and covers key functions such as natural language generation, text-to-image, perceptual model annotation, and model development functions.
SenseTime also demonstrated its ChatGPT-like product, a self-developed Chinese language large model application platform "SenseChat (Chinese: 商量)." As a natural language processing model with hundreds of billions of parameters, SenseChat uses a large amount of data for training and fully considers the Chinese context, which helps it better understand and process Chinese texts. SenseChat demonstrated the ability to understand multiple rounds of dialogue and super-long text at the event. SenseTime also showed several innovative applications supported by the large language model, including a programming assistant, which can help developers write and debug code more efficiently; a health consultation assistant, which provides users with personalized medical advice; a PDF file reading assistant, which quickly extracts and summarize information from complex documents.
At present, SenseTime has created multiple large AI models of CV (Computer Vision), NLP (Natural Language Processing), and AIGC (Artificial Intelligence Content Generation). Its SenseCore AI large device is a rare large-scale model-specific infrastructure in the industry. It has 27,000 GPUs and can output 5,000 PetaFlops computing power. It can perform single-task training with a maximum 4,000-card cluster and can achieve uninterrupted stability for more than 7 days of training.
Based on large AI devices, SenseTime will provide customers with a variety of large models as a service (MaaS, Model-as-a-Service) covering automatic data labeling, large model inference deployment, large model parallel training, large model incremental training, and developer efficiency improvement. SenseTime's SenseNova large-scale model system has fully supported business sectors such as smart cars, smart life, smart commerce, and smart cities and opened up the integration of multiple fields and industries. Application closed loop.
Under the strategic system of "one platform and four pillars," SenseTime's "SenseNova" large-scale model system has fully supported business sectors such as smart cars, smart life, smart commerce, and smart cities and opened up closed-loop applications in multiple fields and industries.
SenseTime saw a 2.2 percent increase in its shares, which closed at HK$3.33 in Hong Kong on Monday.
JD.Com – Will Launch ChatGPT-like Product This Year
On April 8, at the "Artificial Intelligence Large Model Technology Summit Forum" sponsored by the Chinese Association for Artificial Intelligence, He Xiaodong, Vice President of JD.com, disclosed that in response to real industry needs, JD.com plans to release a new generation of large-scale industrial models this year.
In February, JD.com announced that Yanxi (Chinese: 言犀) artificial intelligence application platform under JD Cloud would launch the "JD version" of ChatGPT - ChatJD, with two fields and five applications.
The intelligent human-computer dialogue platform, capable of understanding and generating tasks in natural language processing, has recently emerged as a groundbreaking innovation in retail and finance. This platform boasts five dynamic applications: content generation, human-computer dialogue, user intent understanding, information extraction, and sentiment classification. Its advanced capabilities provide unprecedented accessibility and convenience to users seeking efficient and intuitive solutions to complex tasks in these industries.
The JD large model has been making waves across various industries, focusing on four core areas: text, voice, dialogue, and digital human generation. JD.com's self-research efforts in text generation began in 2019 with the K-PLUG domain model - boasting an impressive 1 billion parameters - to automatically generate product copy of varying lengths for a given stock-keeping unit. These included product titles with a maximum of 10 characters, selling points up to 100, and live-streaming scripts with up to 500 characters. This groundbreaking technology has been extended to cover over 2,000 categories of JD's products, providing customers unparalleled convenience and speed.
At the summit, He Xiaodong also mentioned that the pre-training parameters of Yanxi's large model reached 100 billion levels, covering more than 3,000 categories, the pass rate of manual review exceeded 95%, and the generated text exceeded 3 billion words.
DeepLang AI - Tsinghua University-Backed NLP Startup Reaches Valuation of USD 100Mn
On April 9, QbitAI (Chinese: 量子位) reported that DeepLang AI (Chinese: 深言科技), an AI startup founded by a team from the Natural Language Processing Laboratory at Tsinghua University (THUNLP), has completed a new round of financing with a market valuation of approximately USD 100 million. A new round of financing for DeepLang AI has also begun.
In addition to its focus on large models and strong founder team, there is another reason why DeepLang AI has attracted much attention and soaring valuation recently -it was previously caught in the rumors of acquisition by Wang Huiwen, Meituan (Chinese: 美团) Co-founder. EqualOcrean reported his ambition to invest and create the Chinese ChatGPT rival in February: Meituan Co-Founder Rushes to ChatGPT Trend, Investing USD 50 Million to Start. It is reported that, while finalizing the acquisition of Oneflow (Chinese: 一流科技) and starting the second round of financing, Wang Huiwen is very keen on acquiring two startups incubated from THUNLP: DeepLang AI and ModelBest (Chinese: 面壁智能).
Established in March 2022, DeepLang AI aims to create a new generation of intelligent text information processing engines based on large-scale pre-training models, covering AIGC text generation, information extraction and aggregation, semantic retrieval and other functions, reshaping the entire process of information processing for hundreds of millions of mental workers and tens of millions of information-intensive organizations.
The founder and CEO, Qi Fanchao, is a 2013 undergraduate from the Department of Electronic Engineering of Tsinghua University and a 2017 Ph.D. from the Department of Computer Science and Technology. His main research direction is NLP, and he has published more than 30 papers in top international journals and applied for more than ten invention patents.
Co-founder and COO Li Xiaoxiang is a 2017 Ph.D. from the Department of Electronic Engineering of Tsinghua University. Zhang Han, a partner of Sequoia China, is also one of the company's directors. Sun Maosong, an academician of the European Academy of Sciences, is currently the company's chief scientist.
The current products disclosed by the company include WantWords (Chinese: 反向词典) and WantQuotes (Chinese: 据意查句).
WantWords, a reverse dictionary system that supports looking up words by describing their meanings, exploded on Weibo in November 2021. The so-called "reverse" means that, unlike regular dictionaries, it does not search for meaning by word but instead gives the dictionary a description and lets it help you find words.
In terms of specific operations, enter the meaning you want to express in the search box of the dictionary, and you can get dozens or hundreds of answers. The core AI behind it is called the Multi-channel Reverse Dictionary Model.
In 2019, Qi Fanchao and his classmates cooperated to develop this product, which supports Chinese and Chinese-English cross-language queries and has been open-sourced. Three years later, Qi Fanchao graduated with a Ph.D. and immediately incubated DeepLang AI from THUNLP. The core team all have Ph.D. backgrounds from Tsinghua University. At the same time, the reverse dictionary of the laboratory and the subsequent WantQuotes are also placed under DeepLang AI.
Wang Huiwen, the co-founder of Meituan, has reportedly shown keen interest in procuring two promising startups, DeepLang AI and Model Best, which have been nurtured by THUNLP, an AI incubator. This information was disclosed by AI Tech Talk. If it materializes, the move could be a strategic decision for Meituan, which has been striving to strengthen its AI capabilities and stay ahead of the competition in the fiercely competitive tech industry.