OpenAI, a pioneering artificial-intelligence research organization, has raised the bar once again with the introduction GPT-4 Turbo. This enhanced version of ChatGPT is a significant advancement. OpenAI's latest release continues to push the boundaries of AI, making it more accessible for developers and businesses. GPT-4 Turbo is a powerful language model that has been updated with several new features and updates.
Upgraded database and Token Limits
GPT-4 Turbo's database has been significantly updated, extending the knowledge base to April 2023. The token limit has also been increased from 128 thousand to 128 million tokens. This is equivalent to approximately 300 pages of text. This increased token limit allows users to input longer and more complex texts, allowing them to use a wide range of applications.
Cost Optimization
OpenAI has also optimized the pricing structure to address cost concerns. The price of input tokens dropped by three times, to $0.01 and output tokens have been reduced by two-thirds, to $0.03. GPT-4 Turbo is now a cost-effective option for developers and businesses who want to take advantage of its capabilities.
Updates on Function Calls
GPT-4 Turbo introduces an innovative feature called Function Call Updates. This feature allows users to describe functions of applications or APIs external to the model. It can output JSON object containing arguments that are used for calling these functions. Multiple functions can all be called from a single message. This allows for complex tasks like "opening the car window" and "turning off the air conditioner." GPT-4 Turbo's ability to provide accurate function parameters is a key feature that enhances its practical utility.
Improved Following Instructions with JSON Mode
GPT-4 Turbo excels at tasks that require strict adherence and formatting. It works better in scenarios that demand careful instruction-following, and it also supports a new JSON mode. The response_format API allows the model's output to be restricted to create a JSON object that is syntactically correct. This mode is especially useful for developers who generate JSON outside-of-function calls in the Chat Completions API.
Reproducible Output Data and Logarithmic Probabilities
OpenAI introduces the "initial parameter", which provides reproducible results. This makes it easier to achieve consistent completions when needed. This feature is especially useful for debugging or unit testing. OpenAI plans to release in the near future a function that will return log probabilities of the most likely output tokens produced by GPT-4 Turbo or its predecessor, GPT 3.5 Turbo.
GPT-3.5 Turbo Update
OpenAI also provides an update to GPT-3.5 Turbo, which, by default, supports a 16 KB context window, improved instruction-following, JSON mode, and parallel function invocation. Applications that currently use GPT-3.5 Turbo will be automatically upgraded to the new version on December 11. Older models will be accessible until June 13, 2024.
Assistants API, Code Interpreter, and Search
OpenAI's Assistants API takes AI development to an entirely new level. This powerful tool allows developers to create AI agents for applications. These agents can follow specific instructions and complete a variety of tasks. The Assistants API introduces other features, such as an interpreter to run Python code in an isolated environment and a searching capability to supplement the assistant with domain-specific data. This allows you to interact with external APIs and process files.
API New Features
OpenAI's GPT-4 Turbo supports image inputs through the Chat Completions API. This allows it to read documents with drawings and analyze images, opening up a new world of possibilities for image-related tasks. Developers can integrate DALL*E 3 - an image generation model - directly into their products and applications using the Image API. This includes content moderation built-in to ensure safe and appropriate generation.
Text-to-speech API (TTS), which offers the ability to generate high-quality speech from text is now available. It includes six preset voices, and two variants of the model: TTS-1 or TTS-1 HD. These variants are geared towards real-time use and focus on quality. Pricing starts at $0.015 per 1000 characters.
GPT-4 Fine Tuning: An Experimental Access
OpenAI has launched an experimental access program to fine-tune GPT-4 Custom models, giving organizations a unique opportunity to work with OpenAI researchers to train models in specific domains. These custom models won't be shared with other clients or used to train them, giving organizations exclusive AI capabilities tailored to their needs.
OpenAI's GPT-4 Turbo, along with the updates that accompany it, represent a significant leap forward in AI technology. They provide developers and businesses with the tools necessary to create powerful and innovative AI-driven applications.
These advancements along with the improved token pricing, expanded capabilities, and enhanced capabilities are set to revolutionize the way AI is used across various industries and set new standards for natural-language processing and understanding. OpenAI's commitment remains unwavering as the world of AI evolves. This will pave the way for an AI future that is more powerful, versatile, and accessible than ever before.