OpenAI has made several advancements to its large language models, GPT-4 and GPT-3.5, by incorporating updated knowledge bases and significantly expanding the context window. Additionally, the company has announced that it will take measures to protect its customers against copyright lawsuits, following the examples set by Google and Microsoft.
During its first-ever developer conference, OpenAI revealed that GPT-4 Turbo, which is currently accessible through an API preview, has been trained on data up until April 2023, as opposed to the previous version, GPT-4, which only learned from data up until September 2021. OpenAI plans to release a production-ready Turbo model in the upcoming weeks, although an exact date has not been provided.
One notable improvement in GPT-4 Turbo is the expansion of the context window, which now stands at 128K. According to OpenAI, this extended context window is equivalent to over 300 pages of text contained in a single prompt. Larger context windows enable the language model to better comprehend the question and generate more thoughtful responses. In the past, OpenAI had released two versions of GPT-4, one with a context window of 8K and another with a context window of 32K.
OpenAI has also highlighted the cost-efficiency of GPT-4 Turbo for developers. The input cost per 1,000 tokens has been reduced to $0.01, substantially lower than the $0.03 on GPT-4. Additionally, each output will cost $0.03 per 1,000 tokens. Consequently, OpenAI claims that the latest version of GPT-4 is three times cheaper than its predecessors.
Related Post
Moreover, GPT-4 Turbo supports various types of prompts, including image prompts and text-to-speech requests. It also incorporates DALL-E 3, a feature that was introduced in October. OpenAI emphasizes that GPT-4 Turbo enables users to execute more complex tasks through a single prompt. Users can even specify the coding language of their choice, such as code in XML or JSON, to obtain results.
Furthermore, enhancements have been made to the GPT-3.5 Turbo model, aimed at improving context windows, functionality, and pricing. GPT-3.5 Turbo now features a default context window of 16K and benefits from the same updates as GPT-4 Turbo. The cost for input is set at $0.01, while output incurs a charge of $0.002.
In addition to the advancements made to these models, OpenAI has introduced Copyright Shield, a program that mirrors the copyright indemnity measures offered by Microsoft and Google. OpenAI will now provide copyright protection to enterprise users, covering claims related to copyright infringement. The company assures its customers that it will step in and defend them, including covering the costs associated with legal proceedings.
Copyright Shield will apply to the generally available features of ChatGPT Enterprise and OpenAI’s developer platform.
These recent advancements by OpenAI aim to enhance the performance and accessibility of its large language models. By incorporating updated knowledge bases, expanding the context window, and offering copyright protection, OpenAI strives to provide developers and enterprise users with more effective and efficient language model solutions.
