In a blog post, OpenAI mentioned that the improved GPT-4 Turbo can do things like generating code better than the earlier preview model. The goal is to decrease situations where the model doesn’t finish a task completely. However, the company didn’t provide details about what changes they made.
A few ChatGPT users have recently expressed frustration because the chatbot often declines to finish given tasks. Some of them attributed this to the absence of updates for GPT-4.
However, it’s important to note that the update mentioned is for GPT-4 Turbo, a version of the more commonly used GPT-4. GPT-4 Turbo was trained on information up to April 2023 and is currently in a preview phase. Users of GPT-4, which learned from data available before September 2021, might still encounter similar issues with incomplete tasks.
OpenAI’s GPT-4 Turbo
According to OpenAI’s post, over 70% of users accessing GPT-4 through its API have switched to GPT-4 Turbo due to its more recent and updated knowledge base.
The company mentioned that additional updates for GPT-4 Turbo will be rolled out in the coming months. These updates will include the widespread availability of GPT-4 Turbo with vision. This will enable users to engage in more multimodal tasks, such as generating images from text prompts.
OpenAI has introduced smaller AI models known as embeddings. OpenAI describes embeddings as a “series of numbers that represents the concepts within content, like natural language or codes.”
These assist applications using retrieval-augmented generation, a type of AI that extracts information from a database instead of creating its own response, in understanding the connections between various contents it accesses.
The latest models, namely text-embedding-3-small and a more robust variant known as text-embedding-3-large, are currently accessible.
OpenAI’s GPT-4 Turbo is designed to improve tasks like code generation. Though some ChatGPT users encountered issues, it’s crucial to understand that the updates are for GPT-4 Turbo. A majority of API users have switched, and upcoming features promise more flexibility and accessibility in AI applications.