At its first major developer event, OpenAI announced ChatGPT Turbo on Monday. This is a new large language model (LLM) that promises to address some of the weaknesses of the previous models, GPT-3.5 and GPT-4.
Sam Altman, the company’s top manager, claims that GPT-4 Turbo exceeds the 32K context length that its most advanced enhanced model was capable of supporting until today. GPT-4 Turbo has a huge 128K context length.
GPT-4 is now more capable, it is now “Turbo”.
When we talk about context length we are talking about tokens and each token can be equivalent to a word or phrase. Therefore, the longer the context length, the longer the prompt the model can interpret to give us an answer.
Roughly speaking, because the calculation may vary depending on the prompt, 1,000 tokens are equivalent to about 750 words. So when we talk about a context length of 128K we are talking about more than 300 pages in the case of a standard book, as explained by OpenAI.
OpenAI has also improved the way the model interacts through its API with JSON elements. This is one of the many advanced features that can be used by advanced users who pay for entry tokens to use the tool.
The new GPT-4 Turbo, along with a Turbo version of GPT-3.5 that supports a 16K context window, will be available starting today for firms accessing the API. Recall that a wide variety of applications and services are powered by these models.
Duolingo, Be My Eyes, Stripe, Morgan Stanley, Khan Academy, among others, use GPT-4 in some way or another to make certain features work. Now these and other developers will be able to make use of these new models to enhance their projects.
All of us non-developers will still be able to access GPT-3.5 (with a standard 4.96 token context window) through the free version of ChatGPT and GPT-4 (with the 8.192 token context window) through the paid ChatGPT Plus.
GPT Store, the home of customized GPT models
Although it has not provided too many details, OpenAI has also announced the arrival of customized versions of ChatGPT. These will be created by the community (verified developers in fact) and will be optimized for specific purposes.
These models will be available to ChatGPT Plus and ChatGPT Enterprise users from the end of this month via the GPT Store, a new dedicated platform. There they will find versions of ChatGPT for productivity, education, entertainment, etc.