GPT-4 Turbo Arrives! OpenAI's First Developer Day: AI Text Can Accommodate 300 Pages, Training Exclusive Models Opened

"OpenAI DevDay" - The first development day of OpenAI, the company behind ChatGPT, announced the new product "GPT-4 Turbo" on 10/7. In addition to optimizing existing features of ChatGPT, it also allows AI to read images, speak, and even train custom AI models. In the future, there will also be a GPT store where developers can list their models.

GPT-4 Turbo: Text Capacity Extended to 300 Pages, Faster and Cheaper

GPT-4 Turbo is updated with a database up to April 2023, with a new text window capacity of 128k, allowing for a prompt to display text content exceeding 300 pages. In terms of performance, there have been threefold and twofold efficiency optimizations for input and output.

GPT-4 Turbo will be available for all paid developers to try out via the API, with plans for release in the coming weeks. Additionally, GPT-3.5 will also be updated to the Turbo version.

Assistants API Features

The Assistants API offers new features such as interpreter, retrieval, and function calls to handle previously cumbersome tasks independently, enabling users to build high-quality AI applications.

API New Modes

Visual GPT-4 Turbo

In the Chat Completions API, GPT-4 Turbo can now accept image inputs, allowing for use cases such as generating captions, detailed analysis of real-world images, and reading documents with images, as demonstrated by BeMyEyes, which assists the visually impaired using this technology.

Built-in Image Generation: DALL·E 3, Copyright Protection

Companies like Snap, Coca-Cola, and Shutterstock are utilizing DALL·E 3 for generating images and designs for customers and events programmatically. The new version includes a built-in review feature to prevent misuse: the Copyright shield function, providing built-in protection for users to avoid copyright infringement.

Text-to-Speech

Developers can now generate human-like speech from text using the Text-to-Speech API, with six preset voices to choose from.

Custom AI Models

OpenAI has launched a custom model program that allows selected organizations to collaborate with OpenAI researchers to train a tailored GPT-4 model for their specific domains.

This includes modifying every step of the model training process, from executing additional domain-specific pre-training to running a customized RL post-training process tailored for the specific domain. Organizations will have exclusive access to the custom model.

According to OpenAI's current corporate privacy policy, the custom model will not be provided to other customers or shared with them, nor will it be used to train other models. Additionally, data provided to OpenAI for training the custom model will not be reused in any other context. This will be a highly limited and costly program.Application Link

Comparison of ChaGPT old and new models: