OpenAI’s GPT 4, which can take in up to 25,000 words of text, solve difficult problems with greater accuracy, and accept images as input, is being upgraded to perform even more difficult tasks.
The information was shared as part of OpenAI’s ‘DevDay’ developer conference, with it also revealed that it’s rolling out a platform that would allow users to make custom versions of ChatGPT for specific use cases.
The improved model is dubbed GPT-4 Turbo, and it is currently available via an API preview.
GPT-4 Turbo has been trained with data up until April 2023, compared to the previous version that only learned from data up to September 2021. This means that GPT-4 Turbo has access to more recent and relevant information, which can improve the quality and accuracy of its answers.
Additionally, GPT-4 Turbo will feature a 128K context window, which means users will be able to input more text into the chatbot, equaling roughly 300 pages of text in a single prompt.
Another benefit of GPT-4 Turbo is that it is cheaper to use for developers. The input cost is only $0.01 USD ($0.014 CAD) per 1,000 tokens, compared to the $0.03 USD ($0.041 CAD) on GPT-4. The output cost is $0.03 USD ($0.041 CAD) per 1,000 tokens. Overall, OpenAI claims that GPT-4 Turbo is three times cheaper than the earlier versions of GPT-4.
GPT-4 Turbo is currently available via an API preview. “We plan to release the stable production-ready model in the coming weeks,” wrote OpenAI.
Elsewhere, the company also announced that it will protect its users with a built-in copyright safeguard. The new program is called Copyright Shield, and it aims to protect users from legal claims around copyright infringement.
According to OpenAI, Copyright Shield will cover the generally available features of ChatGPT Enterprise and OpenAI’s developer platform. The company said it will defend its customers and pay the costs incurred if they face legal challenges related to the use of its LLMs.