Analysis of ChatGPT and its impact on the chip industry

ChatGPT is a chat robot program developed by Open Artificial Intelligence Research Center (OpenAI) in U.S. launched?on November 30, 2022. This is a natural language processing tool driven by artificial intelligence technology, using a?neural network architecture?called Transformer, with language comprehension?and text generation.

?

ChatGPT went virus?on the Internet?when it came to the list. Its monthly active users had exceeded 100 million?as of the end of January. None of consuming internet products is faster than ChatGPT to breaking 100 million users. The program’s reference to the new technology RLHF ((Reinforcement Learning with Human Feedback?refers to reinforcement learning based on human feedback) promotes the utilization?of artificial intelligence for?content creation, wring efficiency and extension. Its potential industrialization areas?are Inductive text classification, code development, image generation and intelligent customer service.

?

Recently, Microsoft officially released the ChatGPT version of Bing?and Edge browser, supported by the latest technology from OpenAI. Microsoft claims that the new version?Bing is more intelligent than ChatGPT?and?is able to?chat interactively and help ChatGPT use network knowledge to intelligently connect with OpenAI's technology. Microsoft's stock price closed up more than 4% that day, and its market value soared by more than $80 billion overnight, a record high in five months.

?

As its competitor, Google also came out?a new?product "Bard". This AI chatbot is based on Google's existing language model LaMDA, and it was opened to trustees the day after the launch of the new version of Bing. However, Bard's performance was not really?satisfactory: its wrong answers to objective and factual questions at the press conference not only exacerbated people's concerns that related tools were barely matured?to be integrated into the search engine, but also caused Google's stock price to plummet. As soon as the press conference ended, Google fell by more than 7%, and its market value evaporated by about $102 billion.

?

At the same time, Alibaba is working on research and development. According to a senior technical expert from Alibaba, Dharma Institute?of Alibaba is developing a?dialogue robot?similar to charGPT.?The robot?is currently open to employees in the company for testing. The vice president of JD.com also said that JD.com has substantial?scenarios and high-quality data in the field of ChatGPT?and?its chatJD?will present?the best scenario for the large-scale implementation of AI-Generated Content(AIGC)?and ChatGPT. Baidu is about to?roll out?a resembled?chat AI?in March 2023, and will be the first to apply it to its search service. The name of this project is?ERINE Bot. The firm intends to finish?the internal test in March?before open to the public.

?

However, objectively it is still impossible to achieve a feedback on ChatGPT in China in a short term. The main gap between China and the United States is reflected in hardware and software.?Firstly, the imports and technology of GPU is constrained in China. The gap in hardware results in a huge disparity in computing of Chinese companies versus the American giants. Secondly, there is currently no chat AI comparable to ChatGPT in the world. At present, the companies in China need to spend more than two years on computing power technology to catch up chatGPT. The technology in China is unable to establish a flywheel between calls for real-world data and model iterations, but even if there is such a flywheel, the intelligent level is inadequate.

?

ChatGPT is a product of cross-latitude as a huge revolution in the search area. It will become a next-generation cloud computing service and involves various industries at leading cloud computing companies. Microsoft CEO Nadella announced that ChatGPT will become an open intelligent service of Azure and be fully integrated into other services on January 17, 2023. The long-term trend is optimistic. CICC believes that in the future, large language models like chatGPT are expected to become the most important production tool in the field of AI technology, which will pose new challenges to AI chips supporting high-speed processing of large amounts of data in training and reasoning. There will be the improvement of technologies in chips such as computing power, storage capacity, bandwidth, software and creation of a considerably incremental market space for AI chips.

?

The training and deployment of ChatGPT requires a great deal?of computing power chips. If it focuses on natural language processing tasks, computing in Memory and related chips products are going to be a dominant market. A?data shows that the total computing power consumption of ChatGPT is roughly?3640PF-days. Another inside said that ChatGPT has input?at least 10,000 high-end GPUs?from Nvidia. Citigroup estimates that ChatGPT may drive sales of Nvidia-related products to $3 billion ~?$11 billion within 12 months. The A100 GPU chip released by Nvidia in May 2020 is the largest GPU it has produced so far, consisting of 54 billion transistors,?third-generation Tensor Core. The GPU?has acceleration functions for sparse matrix calculation, which is useful for AI reasoning and training.

?

In addition, Samsung announced in December 2022 that it has developed the world's first large-scale computing system composed by GPU with high-bandwidth memory, processing-in-memory(HBM-PIM) chips. The system is built with 96 GPUs (MI100) from AMD. AMD's MI 100 directly competes against Nvidia's A100 GPU.

?

However, as the attentions to the AI are increasing, it is believed that relevant competing products will gradually appear in the future, and the world may set off another wave of AI research and development. With the promotion of high-performance computing and strong demand for high-performance chips, the overall chip market is expected to see evidently long-term growth in the next few years.