OpenAI has made an announcement regarding the availability of their latest text-generating model, GPT-4, through their API. Existing OpenAI API developers who have a successful payment history can now access GPT-4. The company plans to gradually open up access to new developers by the end of this month, and they will subsequently increase availability limits based on compute availability.
According to OpenAI’s blog post, there has been significant demand for the GPT-4 API, with millions of developers requesting access since March. The usage of GPT-4 for innovative products across various domains is rapidly increasing. OpenAI envisions a future where chat-based models can be utilized for any use case.
Image generated on USP.ai
Compared to its predecessor GPT-3.5, GPT-4 offers enhanced capabilities. It can generate text, including code, and can accept both image and text inputs. In contrast, GPT-3.5 only accepted text inputs. GPT-4 has also achieved a “human level” performance on professional and academic benchmarks. OpenAI trained GPT-4 using publicly available data, including data from public web pages, as well as licensed data.
Currently, the image-understanding capability of GPT-4 is being tested with a single partner, Be My Eyes, and it is not yet available to all OpenAI customers. OpenAI has not specified when it will be expanded to a wider customer base.
It is important to note that, like other generative AI models, GPT-4 is not perfect. It can generate incorrect information and make reasoning errors, sometimes with a high level of confidence. Additionally, it does not learn from its experience and struggles with challenging tasks, such as introducing security vulnerabilities into generated code.
Image generated on USP.ai
OpenAI plans to introduce the ability for developers to fine-tune GPT-4 and GPT-3.5 Turbo, another recent text-generating model, with their own data. This capability, which has been available for several of OpenAI’s other text models, is expected to be released later this year.
Since the unveiling of GPT-4 in March, competition in the generative AI field has intensified. Anthropic, for example, has expanded the context window of their flagship text-generating AI model, Claude, from 9,000 tokens to 100,000 tokens. The context window refers to the amount of text the model considers before generating additional text, while tokens represent the raw text (e.g., the word “fantastic” would be split into the tokens “fan,” “tas,” and “tic”).
Image generated on USP.ai
GPT-4 previously held the record for the largest context window, allowing up to 32,000 tokens. Models with smaller context windows tend to forget the content of recent conversations, often leading to off-topic responses.
In a related announcement, OpenAI has made its DALL-E 2 and Whisper APIs generally available. DALL-E 2 is OpenAI’s image-generating model, while “Whisper” refers to their speech-to-text model. Additionally, OpenAI plans to deprecate older models available through its API to optimize their compute capacity. Due to the increasing demand for generative models, especially ChatGPT, OpenAI has faced challenges in meeting the demand.
Image generated on USP.ai
Starting from January 4, 2024, certain older OpenAI models, specifically GPT-3 and its derivatives, will no longer be accessible. They will be replaced with new “base GPT-3” models that are expected to be more computationally efficient. Developers currently using the old models will need to manually upgrade their integrations by January 4. If developers wish to continue using fine-tuned old models beyond that date, they will need to fine-tune their replacements based on the new base GPT-3 models.
OpenAI assures users that they will provide support to facilitate a smooth transition for those who previously fine-tuned models. In the upcoming weeks, they will reach out to developers who have recently utilized these older models and offer more information once the new completion models are ready for early testing.
