Open AI has announced the general availability of the latest version of your large language model text generation GPT through its payment API. From now on, developers subscribed to the OpenAI paid plan who already have a payment history with OpenAI will be able to access GPT-4.
The company also has plans to open up access to GPT-4 to new developers who sign up for the plan towards the end of this month, to then begin expanding availability limits in the future.”depending on the availability of computing«.
According to those responsible for OpenAI, several million developers have already requested access to the GPT-4 API since last March, «and the range of innovative products that take advantage of GPT-4 grows with each passing day.” Furthermore, they see a future “where chat-based models can support any use case.”«.
GPT-4 can generate text, and even source code, and accepts text and image input. Its predecessor, GPT-3.5 only accepted text. Like all GPT models, it has been trained on data that is publicly available online, as well as data that OpenAI has paid to license.
It’s more advanced, but it hasn’t gotten rid of all the problems of GPT-3.5, as it still “hallucinates” when displaying facts and makes reasoning errors, even showing confidence that things that aren’t are right. Nor does he learn from experience, which makes him fail to solve problems of a certain complexity.
It still doesn’t have all of its features accessible to all GPT users. So, for example, its function of understanding the images that are provided to it as input is not yet available to all OpenAI clients. For now, OpenAI is testing it with only one client, and there are no dates for its opening to the rest of its clients.
In the future, OpenAI will allow developers to tune both GPT-4 and GPT-3.5 Turbo, one of its other newer but less capable models for text generation, with their own data. Apparently they have plans to make it possible before the end of this year.
In addition, OpenAI has taken the opportunity to make other news announcements. Among them the general availability of the DALL-E 2 and Whisper APIs. In addition, they have confirmed that they are going to retire their old models, which are currently available through their API, for «optimize your computing power«. ASs, starting next January 4, 2024, some older OpenAI modelsspecifically GPT-3 and its derivatives, will no longer be available.
They will be replaced by new models based on GPT-3, which will probably be more efficient in terms of consumption of computing resources. Developers using the disappearing templates will need to manually update their integrations by January 4th.