Despite one or two recent blips in usability, OpenAI is pushing on with development of its next-gen chatbot model and has just announced the general availability of GPT-4.
In a blog post, the company announced that all existing OpenAI API developers, who have a “history of successful payments”, will be able to access the latest text-generating model.
Plans to open up GPT-4 access to new developers are in the works, too, and expected to go live by the end of July. This is likely to be followed by OpenAI, who owns ChatGPT, raising availability limits “depending on compute availability”, which it has struggled with recently as the demand for generative models rapidly increases.
Developers Have Been Ready For Quite Some Time
According to the OpenAI blog, millions of developers have been calling for access to the GPT-4 API since March 2023. With the capabilities of the model outweighing anything these devs have used before, it’s little wonder the demand has grown.
The system’s predecessor, GPT-3.5, only accepted text, so developers are understandably keen to explore the possibilities of a model that can generate code, and accept image and text inputs.
However, before you get too excited it’s worth knowing that the image-understanding functionality won’t be available right away. This is still in testing with the AI-powered virtual assistance app Be My Eyes. It’s not yet known when this capability will go live.
Much like previous GPT models, GPT-4 has been trained using publicly available data, such as content from web pages, as well as data licensed by OpenAI itself.
It’s All Hands on Deck For Refining The Models
With reports of ChatGPT’s fake news seemingly surfacing every other day, we know the system isn’t perfect. As well as being unable to introduce security vulnerabilities into the code it creates, it’s becoming well known for its hallucinations and information errors.
OpenAI is well aware of these hindrances and has said it will allow developers to “fine-tune” GPT-4 and GPT-3.5 Turbo – one of the original models that powers ChatGPT – with their own data. The caveats are, they must have been able to do it with several other of the system’s text-generating models, and they won’t be able to get hands-on until later this year.
The APIs keep Coming, But Older Models Will Disappear
Not content with making GPT generally available, OpenAI have also opened up the APIs for DALL-E 2 – the company’s image-generating model – and Whisper – its speech-to-text model.
However, you can expect to see older models of the system – such as GPT-3 – becoming unavailable. It’s understood they can’t keep up with computer capacity and will be replaced with a new, more efficient “base GPT-3” model. OpenAI has stated it’ll provide developer support with the transition.
“In the coming weeks, we will reach out to developers who have recently used these older models, and will provide more information once the new completion models are ready for early testing,” OpenAI said in a statement.
Developers will have until January 4th 2024 to manually upgrade their old models.
