OpenAI Roles Out Big Upgrades
OpenAI is rolling out some significant upgrades to their API!
Created on November 7|Last edited on November 7
Comment
OpenAI has announced a series of updates and expansions to their artificial intelligence platform, marked by significant improvements in capabilities and reductions in costs. The introduction of the new GPT-4 Turbo model stands out as a major enhancement, boasting a substantial 128K context window, which allows it to process the equivalent of over 300 pages of text at once. This new model also prides itself on being three times less expensive for input tokens and twice as less for output tokens when compared to its predecessor, GPT-4.
3.5+
In addition to GPT-4 Turbo, the platform has launched a refined version of GPT-3.5 Turbo with a 16K context window and enhanced functionality in terms of instruction following and JSON format responses. Improvements in the function calling feature are also notable, enabling more accurate and diverse application interactions in a single request.
Assistants API
For developers, the Assistants API is a new introduction aimed at simplifying the creation of AI-driven applications. This tool is designed to manage complex tasks and can leverage additional tools such as the Code Interpreter and Retrieval functions to perform advanced operations. Assistants can now handle persistent threads, thus bypassing previous context window limitations.
OpenAI also highlights the upcoming ability to generate reproducible outputs, which will aid in debugging and testing, and the feature to view log probabilities for generated tokens.
Vision API
As for multimodal capabilities, GPT-4 Turbo has expanded to include vision, enabling image analysis and detailed interpretation in the Chat Completions API. DALL·E 3, the image creation tool, is now accessible to developers, and a new text-to-speech API offers high-quality voice generation from text.
GPT-4 Fine Tuning
An experimental program for GPT-4 fine-tuning is being established, though preliminary findings suggest that fine-tuning GPT-4 might not provide as significant improvements as it did with GPT-3.5. However, a Custom Models program is also launching, allowing organizations to train GPT-4 on large, domain-specific datasets for bespoke solutions.
It's encouraging to see OpenAI's dedication to affordability and accessibility in AI technology. OpenAI's efforts to lower costs associated with its GPT-4 API is a commendable step that makes this cutting-edge technology more accessible to a wider range of developers and companies. By doing so, OpenAI is enabling more innovators to create and integrate advanced AI applications, which will likely lead to many more AI-assisted products and services.
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.