OpenAI upgrades GPT-4 Turbo for paid ChatGPT users, "improved writing, math, coding"

OpenAI has announced an update to its GPT-4 Turbo language model, targeted towards paid users of its ChatGPT platform. This new iteration of GPT-4 Turbo shows improvements in several areas, including writing, mathematical capabilities, logical reasoning, and code generation.

One of the most noticeable changes for users will be the way ChatGPT interacts through text. GPT-4 Turbo is designed to deliver more concise and conversational responses. This is achieved through a combination of factors, including improved training data and a focus on generating direct and clear language.

The upgraded model has access to a broader knowledge base, having been trained on information up to December 2023, compared to the previous version’s data cutoff of April 2023.

Initial feedback from users who have trialed GPT-4 Turbo suggests promising results. In the area of coding, for instance, one user reported a “big jump in math/reasoning” during benchmark testing.

Another user commented on the model’s ability to generate more streamlined code while including details, such as automatically adding a flag to download high-quality videos.

A separate user noted the model’s improved efficiency, achieving the same level of performance with 64,000 tokens (units of text used by the model) as the previous version required 26,000 tokens.

Currently, this new GPT-4 Turbo is only accessible to paid subscribers of the ChatGPT platform. These include users on ChatGPT Plus, Team, and Enterprise plans, along with those who interact with the model through OpenAI’s application programming interface (API), released a few days ago.

More here.