Per editor: Du Yu
On July 7, OpenAI announced on its official website that the GPT-4 API is fully open for use. All paid API users now have direct access to GPT-4 in 8K context without any waiting.
Image source: OpenAI official website
GPT-4 API, fully open for use
OpenAI says the GPT-4 is our most capable model. Since March, millions of developers have requested access to the GPT-4 API, and the range of innovative products that leverage GPT-4 is growing every day.
Today, developers around the world can use GPT-4 language models to enhance their own applications or develop entirely new generative AI applications. We plan to open it to new developers by the end of the month, and then start raising the rate limit based on compute availability.
Now that it's officially fully open, developers around the world can use GPT-4 language models to enhance their own applications or develop new generative AI applications.
Image source: OpenAI official website
OpenAI says we launched the Chat Completion API in March, and it now accounts for 97% of our API GPT usage.
The original Completions API, launched in June 2020, provided free-text prompts for interacting with our language models. Since then, we've learned that using a more structured prompt interface often provides better results. The chat-based paradigm has proven to be very powerful, handling the vast majority of previous use cases and new conversational needs while providing greater flexibility and specificity. In particular, the structured interface of the Chat Completion API (e.g., system messages, function calls) and multi-turn dialogue capabilities enable developers to build conversational experiences and a wide range of completion tasks. It also helps reduce the risk of instant injection attacks, as the content provided to users can be structurally separated from the instructions.
OpenAI said that based on the stability of these models and the readiness of production-scale use, we have also launched GPT-3.5 Turbo, DALL· E and Whisper API. We are safely enabling fine-tuning for GPT-4 and GPT-3.5 Turbo, which is expected to be available later this year.
In addition, OpenAI has announced a number of deprecated models and completely new versions. For example, davinci-instruct-beta, gpt-curie-instruct-beta, text-ada-001, text-babbage-001, text-curie-001, text-davinci-001, text-davinci-002 and text-davinci-003 replaced by GPT-3.5-turbo-instruct.
GPT-4 is more creative than ever
GPT-4 is more creative than ever. You can generate, edit, and iterate creative and technical writing tasks with users, such as creating songs, writing screenplays, or learning users' writing styles to generate content.
GPT-4 can accept pictures as input and generate descriptions, classifications, and analysis. For example, enter an image of eggs, flour, and add the sentence "What can I do with these ingredients?" ”
ChatGPT answers: You can make many choices with these ingredients: pancakes or waffles, crepes, French toast, omelets or omelets, quiche custard or pudding, cakes or cupcakes, muffins or bread, biscuits.
GPT-4 is capable of processing text over 25,000 words, allowing for use cases such as long-form content creation, extended dialogue, and document search and analysis. For example, you can add a link directly to the content of the question.
ChatGPT's popularity has continued to decline recently
The popularity of ChatGPT, which has been popular since November 2022, has declined significantly in recent times. According to SimilarWeb, in the first five months of 2023, ChatGPT global visits increased by 131.6%, 62.5%, 55.8%, 12.6% and 2.8% month-on-month, respectively, while the June data fell by 9.7% month-on-month, showing negative growth for the first time.
According to SimilarWeb, ChatGPT has exceeded the 100 million active user mark since its launch in the United States in November 2022. The traffic explosion at the beginning of the year was mainly driven by users on the ToC side, and ChatGPT's user experience far surpassed the previous generation gave users a great sense of freshness, and spread quickly through social media such as TikTok. According to the Google Popularity Index, the popularity of "AI" keywords increased by 267% from October 2022 to April 2023. But then, starting in May, the growth in popularity of "AI" keywords stalled, and the week from May 14 to May 21 saw its biggest decline in nearly a year.
According to the analysis of First Finance and Economics, CITIC Securities believes that the decline in ChatGPT user traffic is mainly due to the decrease in the freshness of individual users, as well as OpenAI's active adjustment of operation strategies (computing power resource cost constraints), and the focus on the 2B field, etc., the most intuitive evidence is: at present, ChatGPT-related plug-ins and development tool chains continue to maintain a high degree of popularity, while most European and American software enterprise products have begun to fully integrate generative AI.
In the long term, the big language model has a natural advantage in improving production efficiency, the application scenarios of the relative segmentation of the enterprise market, superimposed on the unique scenarios & business data of the enterprise, etc., are expected to make the development process and market space of the big language model in the 2B field significantly better than the 2C market, and the Copilot proposed by Microsoft will be the most feasible landing method. Continue to be optimistic about the development process of the global AI industry, and suggest that in the second half of the year, we should focus on the progress of US software companies in the field of AI and corresponding investment opportunities.
Daily economic news comprehensive OpenAI official website, CBN
Daily economic news