Share
On November 6th, OpenAI organized a Developer Day, where Sam Altman, the CEO of OpenAI, introduced a series of new models and features that the company plans to officially launch in the near future. These announcements not only attracted AI developers but also the general public. Let's take a look at the new models shared by the CEO of OpenAI.
GPT-4 Turbo is the newest upgraded version from OpenAI, and it comes with notable features such as:
GPT-4 Turbo has been enhanced compared to the GPT-4 version.
Increased Context Length up to 128,000 Tokens. The significant boost in context length compared to GPT-4, which was 8,000 and 32,000 (in some cases), "equivalent to more than 300 pages of text in a single query," the CEO shared.
It was updated with additional data about world events. While the previous version only updated until the end of 2021, GPT-4 Turbo will include updates until April 2023. Sam also affirmed that they will continue to develop to provide data closer to the present.
The text-to-speech feature has been updated. Developers can switch between 6 different voices through the API naturally and expressively like humans.
GPT-4 Turbo is more cost-effective than GPT-4. GPT-4 Turbo has a cost that is only 1/3 of GPT-4 for input tokens (0.01 USD for 1000 tokens) and about 1/2 the cost of GPT-4 for output tokens (Approximately 0.03 USD for 1000 tokens).
Allows images as input for conversation in the Chat Completions API for purposes such as reading images, creating captions, detailed analysis, or reading documents and data. This feature is expected to be developed by Open API into one of the main modules of GPT-4 Turbo in the upcoming release.
At this event, Open AI also introduced the "Assistants API" with the aim of creating a personal assistant for developers during their application development. The Assistant API offers features such as a Code Interpreter, Retrieval, and function calls to assist in handling time-consuming tasks that engineers and developers had to do themselves before.
For organizations and businesses with high demands for suitability and accuracy tailored to their needs, Open AI has launched the Custom Model Program, a customization program. With this program, business organizations (especially those with extremely large exclusive datasets) will collaborate with research experts at Open AI to train a custom GPT-4 model to serve their specific needs. With this program, the data provided by organizations and businesses will be served exclusively, and Open AI will ensure data security policies for them.
However, Open AI notes that this program will have limitations, and they caution that the costs may be significant.
GPTs is a customized version of Chat GPT designed for specific purposes. Users can provide instructions, knowledge, or specific actions to build their own version of Chat GPT. Business customers can use GPTs to create an in-house chatbot based on the company's proprietary knowledge data.
ChatGPT can now be specialized in topics through GPTs.
DALL·E is OpenAI's text-to-image model. With the latest version, DALL·E 3, developers can integrate the DALL·E 3 model into ChatGPT Plus and enterprise user environments.
DALL·E 3's API can now be integrated into ChatGPT Plus and Enterprise User
Expected later this month, OpenAI will launch GPT - Store, a platform where developers, following the company's policies, can publish their own GPT products. OpenAI will pay developers if their GPT is widely used.
With the announcements at the developer conference, OpenAI is increasingly asserting its leading position in the AI field with a series of improvements and new model ideas. This is considered a goldmine for AI developers to explore. Furthermore, the upcoming groundbreaking models could usher the world into a new era of artificial intelligence, posing many questions about the relationship between humans and AI and how we "coexist" with it.
Share