The Zhitong Finance App learned that Guotai Junan released a research report saying that OpenAI launched GPT-4 Turbo, which is more cost-effective, fine-tuning the API has also achieved model customization, and the developer ecosystem has been further improved.The AI application store has ushered in a new era.Recommended targets: iFLYTEK (002230.SZ), etc., beneficiary targets: Wanxing Technology (300624.SZ), etc.

▍ Guotai Junan's main views are as follows:

The GPT-4 Turbo makes ChatGPT more convenient to use, with significant performance improvements.

ChatGPT using GPT4 Turbo can integrate functions such as writing and running code, analyzing data, and generating images without the need for drop-down menus like GPT4, making ChatGPT truly a proxy. OpenAI's new GPT-4 Turbo supports 128,000 tokens, and the context length is up to 32K. In addition, the JSON model ensures that the model responds effectively, makes it easier to call the API, and realizes a higher degree of control over the model.

Turbo Vision, DALL.E3, and a new text-to-speech model are all embedded in the API. GPT 4 Turbo has world knowledge as of April 2023 and can bring knowledge from external documents or databases into the constructed content.

Thanks to the fine-tuning API, everyone can customize their own GPT.

The fine-tuning API aims to achieve better performance and model customization in various applications with relatively small amounts of data. Its functions include modifying every step of the model training process, performing additional pre-training in specific fields, tailoring the RL training process for specific fields, and uploading documents in self-constructed GPT.

Based on this, OpenAI has also launched a GPT store. Users can build private GPT to use, and can also share it to the app store by providing a link. Developers who build the most practical and commonly used GPT will also receive a certain amount of revenue.

GPT-4 Turbo speeds up and reduces fees, and introduces a “copyright shield” mechanism.

Compared to GPT4, while doubling the number of tokens per minute, the GPT-4 Turbo also reduced the API rental price. Prompt tokens are 3 times cheaper, and completion tokens are 2 times cheaper. Furthermore, in order to ensure the safe construction of applications on the platform, a copyright shield mechanism has been introduced, which means that if faced with legal claims relating to copyright infringement, OpenAI will step in to protect customers and pay the costs incurred. At the same time, OpenAI also promised not to conduct training through APIs or enterprise data.

It can be seen from this that, from technical performance to commercial mechanisms, the AI application development ecosystem has been fully improved.

Risk warning:

Technology iterations fell short of expectations, and the pace of AI application market expansion fell short of expectations.