Recently, under the upsurge of ChatGPT, several companies have officially announced that they will do ChatGPT or related news. Among them, some companies have indeed been deeply involved in the field of AI for a long time, and have certain technical accumulation; some companies also took the opportunity to hype And playing gimmicks, hoping to temporarily increase the stock price or divert attention (often due to poor management).
Starting from 2020, the industry believes that the AI industry has entered a bubble period, and the specific implementation has caused many companies that are not deep in technology and are burning money. Against this background, the rise of ChatGPT has given the entire AI industry a glimmer of hope. Even though smarter virtual assistants have appeared before, the breadth of knowledge reserves and flexible response levels demonstrated by ChatGPT make all smart assistants ashamed, and there are already countless future application scenarios worth looking forward to.
Eager domestic enterprises
From the Internet to technology to finance, Chinese companies preparing to join this battle have already made every effort, not only for the stock price, but also for the possibility of having a story to tell in the future.
baidu
Baidu is building a Chinese version of the ChatGPT-like project-ERNIE Bot. Baidu has a full-stack layout in the four-tier architecture of artificial intelligence. Including the underlying chips, deep learning frameworks, large models, and top-level search applications. And Wenxinyiyan is located at the model layer. Baidu said that the product is currently preparing for the sprint before it goes online. It is expected to complete the internal test in March and open it to the public.
Ali
A reporter learned from Alibaba that the Alibaba version of the chat robot ChatGPT is under development and is currently in the internal testing stage. Previously, there were media reports that Alibaba Dharma Academy was developing a ChatGPT-like dialogue robot. Judging from the exposed screenshots, Alibaba may deeply combine AI large-scale model technology with DingTalk productivity tools. It is reported that the dialogue robot of Ali’s version of ChatGPT will be deeply integrated with DingTalk.
Millet
Xiaomi said: There are rich landing scenarios in the ChatGPT field, and it will increase human and resource investment in related fields in the future.
According to people familiar with the matter, Xiaomi has rich implementation scenarios in the field of ChatGPT, including Xiaoai dialogue, robots, etc. Among them, Xiaoai has huge data support, which will allow Xiaomi to make faster progress in large-scale models. It is reported that Xiaomi has made multiple parallel attempts on AI large models before, and will increase human and resource investment in related fields in the future.
Jingdong
JD.com plans to integrate ChatGPT’s methods and technical points into product services to promote the industrialization of artificial intelligence. He Xiaodong, vice president of JD Group and IEEE Fellow, said that JD.com is the best scenario to accelerate the large-scale implementation of AIGC & ChatGPT. He said that JD Cloud’s Yanxi artificial intelligence platform has focused on text, voice, dialogue and digital human generation, and will use ChatGPT and other related technological achievements to accelerate the application of artificial intelligence in my country. He Xiaodong introduced that JD.com has carried out related work in terms of text generation, speech generation, dialogue generation, and digital human generation.
NetEase Youdao
Youdao AI technology team has invested in the research and development of ChatGPT homologous technology (AIGC) in educational scenarios. At present, the team has tried to apply it in subdivided learning scenarios such as AI oral teachers and Chinese composition correction. The company is expected to launch related demo products as soon as possible. The launch of this product will mark the first landing of AIGC technology in the domestic Internet education scene.
Tencent
Tencent said that at present, there are layouts in relevant directions, and special research is also advancing in an orderly manner. Tencent continues to invest in the research and development of AI and other cutting-edge technologies. Based on its previous technical reserves in the fields of AI large models, machine learning algorithms, and NLP, it will further carry out cutting-edge research and application exploration.
According to insiders, on the product side, Tencent has launched a Hunyuan AI large model, which completely covers basic models such as NLP (Natural Language Processing), CV (Computer Vision), multimodality, and many industry/domain models. Tencent said that with the rise of AIGC, in the future, the Hunyuan AI large model will continue to promote continuous upgrades in the fields of text content generation and Vincent graphs.
360
On February 9th, 360 pointed out in the announcement of abnormal fluctuations in stock trading that the company currently has continuous R&D and computing power investment in AIGC technologies such as ChatGPT, text generation and image technologies, but so far, all the formed The results are only used as a productivity tool for the company’s internal use. The indicators of the company’s ChatGPT-like technology can only reach a level slightly stronger than GPT-2, and there is still a generation gap compared with the current ChatGPT.
On the same day, Zhou Hongyi, the founder of 360, talked about ChatGPT for the first time when he had a conversation with Zhang Chaoyang, the founder of Sohu. He bluntly said that if an enterprise cannot take the bus of ChatGPT, it is likely to be eliminated. At the same time, he pointed out, “Companies that are search engines have the best opportunity to participate in the research of ChatGPT-like products, and 360 will definitely not give up the research and tracking of this technology.”
Tuya Smart
It is understood that Tuya Smart, a global IoT development platform service provider, will soon launch ChatGPT applications in smart home and smart business scenarios, and will release a demo in the near future.
Tuya’s technical team has initially completed the integration and expansion of ChatGPT and its own AI model. By introducing ChatGTP’s powerful language processing capabilities, it will further enhance the digital level of smart homes and vertical industries. It is worth noting that the new application of Tuya may be connected to the Bard dialogue service released by Google based on the LaMDA large model.
blue cursor
BlueFocus said on the interactive platform that the company is actively trying ChatGPT technology. Its own virtual person Su Xiaomei has participated in multiple rounds of conversations with AI. The ability of AI poetry writing has entered the technical testing stage. In the future, it will combine AIGC and terminal interaction to do more digital expansion. ; At the same time, the company is also doing trial and debugging of other ChatGPT products.
HKUST Xunfei
HKUST Xunfei said that ChatGPT mainly involves natural language processing related technologies, which is one of the applications in the field of cognitive intelligence. The company has long-term and profound accumulation in technology and applications in this direction. HKUST Xunfei Al learning machine will be the first product of the company’s ChatGPT-like technology, and it will be released in May this year.
Qi Anxin
Qi Anxin Artificial Intelligence Research Institute is training Qi Anxin’s proprietary ChatGPT-like security large model based on ChatGPT related technologies and its own security knowledge and data. In the future, it will be widely used in security product development, threat detection and other fields.
Bank of Jiangsu
The bank has tried to use ChatGPT technology to improve software development productivity and further improve the efficiency of technology operations. “From the test results, the code is running in the production environment, and it takes less than an hour to perfectly complete all the requirements. Not only the time for writing functions is greatly shortened, but also the time spent on communication with manufacturers has been shortened from a few days to a few days. hours.” Said the relevant person in charge of the Information Technology Department of Bank of Jiangsu. “In the future, software development may develop into a model written by AI and reviewed by a dedicated person, achieving a substantial leap in productivity.”
The relevant person in charge said that the bank will further integrate the ChatGPT new technology into daily office work, risk assessment, customer service, auxiliary marketing and other fields, so as to meet the financial needs of customers in different scenarios faster.
In addition to the above-mentioned companies, Hanwang, Secoo, Shuidi, Entropy Technology, Montnets Technology, Tiandi Online, Jiachuang Video, Fantuo Digital Technology, etc. have also officially announced plans to make efforts on ChatGPT, but more of them are rubbish. Hotspots do not have much substantial technical accumulation.
build localizationChatGPTnecessity and feasibility
Li Yanhong once commented on ChatGPT and AIGC at the internal conference in early 2023: “Technology can achieve this step, but what kind of product it will become and what kind of demand the product can meet, there are still many uncertainties in this chain .”
According to the research report of Guosheng Securities, behind the intelligent and smooth communication experience of ChatGPT, it has as many as 175 billion model parameters. The public crawler data set mainly used by OpenAI has a human language data set of more than one trillion words. In terms of computing power, GPT-3.5 is trained on Microsoft Azure AI supercomputing infrastructure (a high-bandwidth cluster composed of Nvidia V100 GPUs), with a total computing power consumption of 1 trillion calculations per second, running about 3640 full days .
In terms of cost, Li Di, CEO of Xiaoice, gave an example: If the ChatGPT method is used, the daily cost will be as high as 300 million yuan, and the annual cost will exceed 100 billion yuan, based on the amount of dialogue interaction currently supported by the Xiaoice framework.
Even so, the industry still generally believes that China should also have its own ChatGPT. It is not to speculate on concepts, but to truly practice related technologies and iterative models. Startubit CEO Zhang Jiong believes that China needs to build its own ChatGPT in three aspects:
“The first is the market side. At present, ChatGPT is not open to domestic users. Domestic users can’t use it as they like overseas users. From the market level, there will inevitably be a Chinese ChatGPT. The second is the technology side, which can be seen in the market at present. The large-scale model (LLM) such as ChatGPT’s basic model GPT-3.5 is basically trained in English as the native language. The effect of the Chinese-based natural language processing (NLP) is still slightly inferior to the English-based model, so there is a Chinese-based model. It will further improve the user effect. The third is data security. The content generation of AI itself is the result of training based on a large amount of data, and OpenAI has gradually changed from a public welfare project to a market-oriented project. There may be uncertainties in the future due to factors such as interests In addition, mainland China requires all data to be stored locally, and OpenAI does not have a team in China, so it is difficult for localized storage and maintenance of data to meet regulatory requirements.”
Especially the third aspect. In theory, ChatGPT will not take the initiative to train new data. Given the process of training data, ChatGPT may answer with racial discrimination, sexism or other inappropriate remarks. If someone deliberately uses a product like ChatGPT There are potential social risks in interfering with public opinion and provoking social sentiment.
If we do not have our own ChatGPT, we will not be able to control the owner and type of data used to train the model, and will not be able to correct it later through algorithms and training data. The existence of a large amount of false data will also bring a burden to the accountability of intellectual property rights and social issues. Came extremely difficult.
In the future, if AI technology is popularized on a large scale, and the generation gap between us and foreign AI large models widens to the point where we have to use it, then ChatGPT will also face a similar chip bottleneck. In order not to fall into a passive situation and continue to enjoy the benefits brought about by the technological upgrade of the AI industry, it is necessary to build ChatGPT in China.
Regarding the difficulty of China’s self-developed ChatGPT products, Zhang Jiong told Dongdian Technology: “Currently, the chatbots used on a large scale in China are related to the sub-tasks of NLP, such as machine translation, intelligent customer service, question answering systems, etc. With the development of LLM China will also popularize LLM-based AI chatbots.”
He believes that China’s self-developed chat robots need to meet at least the following requirements in terms of application: First, at the interaction level, robots need to adapt to the expressions of Chinese habits, that is to say, they must understand the commands expressed by humans. In addition, in the communication stage, in order to achieve barrier-free communication, the robot needs to be familiar with relevant knowledge such as Chinese history and culture, and use Chinese language style and expression habits to communicate, such as the same word expressing different meanings and emotions in different contexts. Furthermore, chatbots need to provide personalized services based on Chinese user needs and social status, such as Chinese-specific payment methods and ethnic customs.
Since domestic investment in AI technology, especially large-scale models and related chip computing power, and the underlying technology are far behind that of the United States, it is not easy to reach the level of OpenAI in the short term.
However, in the long run, China has excellent algorithm engineers, a unified large market, rich application scenarios and data sources, and Zhang Jiong also bluntly said that it is not impossible to do a good job in China’s ChatGPT. “China has gone a long way in terms of big data. Various domestic scenarios are highly informatized, with rich data sources and a complete related industrial chain. However, the model is slightly insufficient. At present, the basic GPT-3.5 of ChatGPT These are all large-scale models. Large-scale models have large initial investment and slow results. In the past, domestic investors were not the preferred targets. Therefore, only a few large Internet companies will participate, but the investment is limited and the progress is slow. However, This is also an opportunity, I believe that this wave of ChatGPT is a warning to domestic industrial investors and Internet companies, and there should be greater investment in the future.”
