OpenAI has recently made its latest text-generating model, GPT-4, available via its API. This exciting news means that developers who have a successful payments history with OpenAI can access GPT-4 immediately, while new developers will gain access by the end of the month. OpenAI plans to gradually increase availability limits based on compute availability.
The demand for GPT-4 has been overwhelming, with millions of developers requesting access to the API since March. OpenAI is delighted to see the range of innovative products leveraging GPT-4 growing every day. Their vision is to have chat-based models that support any use case.
The capabilities of GPT-4 go beyond its predecessor, GPT-3.5. GPT-4 is not only able to generate text, including code, but it can also accept image and text inputs. It performs at a human level on various professional and academic benchmarks. To train GPT-4, OpenAI utilized public web pages and licensed data, as they did with their previous GPT models.
While GPT-4 is already impressive, OpenAI has acknowledged that it has its flaws. The model sometimes hallucinates facts and reasons incorrectly, displaying a misplaced sense of confidence. Additionally, it fails to learn from its mistakes, which introduces security vulnerabilities into the code it generates.
OpenAI has plans to allow developers to fine-tune GPT-4 and another text-generating model called GPT-3.5 Turbo with their own data. This feature is expected to be available later this year.
Competition in the field of generative AI has intensified since GPT-4’s initial reveal in March. For example, Anthropic has increased Claude’s context window from 9,000 to 100,000 tokens. GPT-4 previously held the title for the largest context window at 32,000 tokens. Models with small context windows tend to forget recent conversations and veer off-topic.
In addition to the release of GPT-4, OpenAI has also made its DALL-E 2 and Whisper APIs available. DALL-E 2 is a model that generates images, while Whisper is a speech-to-text model. OpenAI plans to deprecate API models to optimize their compute capacity as they have been struggling to meet the demand for generative models due to the popularity of ChatGPT.
OpenAI has announced their intention to replace GPT-3 and its derivatives with base GPT-3 models on January 4, 2024. These new models are expected to be more compute efficient. Developers currently using the old models must manually upgrade their integrations by the specified date. Fine-tuned old models can still be used on top of the new base GPT-3 models, and OpenAI will provide support to ensure a smooth transition.
This update from OpenAI brings exciting possibilities for developers and users alike. GPT-4 and its enhanced capabilities promise to unlock new opportunities for generating text, code, and more. As OpenAI continues to refine their models and address their limitations, the potential for generative AI only continues to grow.
