Recently, OPENAI Operations Co., LtdOPCO, LLC) has applied for the registration of two "GPT-5" trademarks in China, including scientific instruments and website services under international classification. The current trademark status is currently under application. In addition, according to media reports, a document disclosed by the United States Patent and Trademark Office (USPTO) shows that OpenAI submitted a trademark application for GPT-5 on July 18th. And it has been received. In this disclosed application, OpenAI mentioned that the yet to be released GPT-5 will have many capabilities that GPT-4 does not have, and almost every one points to AGI.

01 | GPT-5 is coming soon? OpenAI mentioned the basic information of GPT-5 in this trademark application. GPT-5 will be a Large Language Model (LLM) for natural language processing, generation, comprehension, and analysis. It is worth noting that in addition to text, GPT-5 will also have the ability to understand and generate speech. Specifically, the functions of GPT-5 include translating text or speech from one language to another, speech recognition, and generating text and speech. In addition, GPT-5 may also have the ability to learn, analyze, classify, and respond to data. However, it should be noted that although a trademark application has been submitted, this does not mean that GPT-5 will be developed immediately.

The specific functions and improvements of 02 | GPT-5 still need to be officially confirmed by OpenAI. According to OpenAI's previous plan, GPT-4.5 will be released to the public in September or October this year, and then GPT-5 will be released before the end of the year. Now that August has entered, there is not much left on OpenAI's schedule. In May, members of the OpenAI development team also tweeted that GPT-5 may achieve ArtificialGeneral Intelligence (AGI), which is an artificial intelligence with human intelligence and the ability to complete various tasks. However, it should be noted that in early June, OpenAI CEO Altman revealed in a speech that the company had not started training GPT-5. As a reference, OpenAI spent 6 months conducting adversarial testing before releasing GPT-4 to ensure the security of the AI as much as possible (the company did not disclose details such as the training duration of the model itself). Considering this, we may have to wait at least a few months to see GPT-5. The trademark application of OpenAI is only to prevent others from using the name GPT-5, and does not mean that the development of the large model has been completed.

03 | GPT-5 will have some new features including multimodal integration with industry discussions, and GPT-5 will have some new features including multimodal integration. At present, GPT-4 can only handle two types of data: text and image. People hope that GPT-5 can make breakthroughs in other modes such as voice and video. In addition, combined with the update trend of GPT-4, people predict that GPT-5 will develop towards a software as a service (SaaS) platform. GPT-5 will not only be a software, but can be integrated into various websites, mobile applications, and computer systems, and have multiple vertical domain plugins. However, the training of GPT-5 will consume a larger scale of computing power. An article predicts that GPT-5 may require 30000 to 50000 integrated H100 graphics cards. People predict that GPT-4 may have been trained on 10000 to 25000 A100 sheets. Meta has approximately 21000 A100s, Tesla has approximately 7000 A100s, StabilityAI has approximately 5000 A100s, and Falcon-40B trained on 384 A100s. It is not difficult to predict that by the time OpenAI releases GPT-5, it will also make progress in this field, making it more useful in various work areas, not just as a chat robot or AI image generator.

04 Jingtai Viewpoint | ChatGPT's recent popularity has continued to decline. Since November 2022, the continuous popularity of ChatGPT has shown a significant decline in recent times. According to data from SimilarWeb, the global traffic growth of ChatGPT in the first five months of 2023 was 131.6%, 62.5%, 55.8%, 12.6%, and 2.8% month on month, respectively. However, in June, the data showed a 9.7% month on month decrease, marking the first negative growth. The decline in ChatGPT user traffic is mainly due to a decrease in personal user freshness, as well as OpenAI's proactive adjustment of operational strategies (constrained by computing resource costs) and shift focus to the 2B field. The most direct evidence is that ChatGPT related plugins and development tool chains continue to maintain high popularity, and most European and American software enterprise products are beginning to fully integrate generative AI. In the long-term dimension, the big language model has natural advantages in improving production efficiency. The relatively segmented application scenarios of the enterprise market, combined with the unique scenarios and business data of each enterprise, are expected to make the development process and market space of the big language model in the 2B field significantly superior to the 2C market. Microsoft's proposed Copilot will be the most feasible landing method. Continuously optimistic about the development process of the global AI industry, and it is recommended to focus on the progress and corresponding investment opportunities of American software companies in the AI field in the second half of the year.