The global tech industry is witnessing a milestone as OpenAI’s much-anticipated GPT-4 API becomes broadly available to all paying customers.
Since its launch in March, the GPT-4 API has been gaining traction with developers around the world, but until now it has only been available to a select group of customers. Now, these developers will have the opportunity to harness the power of GPT-4, which is currently being hailed as the most proficient model available.
GPT-4 for Everyone
The GPT-4 API provides an 8K context window, which refers to the amount of text information that the model can "consider" or "remember" when generating a response. This 8K context window essentially means that the model considers the last 8,000 tokens (roughly 8,000 words or characters, depending on the language) to generate its output, and this feature is critical to keeping the model's responses consistent and coherent.
The GPT-4 API is accessible to existing API developers with a history of successful payments. OpenAI plans to expand access to new developers by the end of the month, based on compute availability, and then increase the rate limit.
Other API versions and development
The company also made GPT-3.5 Turbo, DALL·E (image generation), and Whisper (audio) APIs generally available, indicating that these models are ready for production-scale use. In addition, OpenAI is working on fine-tuning GPT-4 and GPT-3.5 Turbo, and developers can expect this feature later this year.
In the context of AI models, fine-tuning refers to taking a pre-trained model (one that has already learned general patterns from a large dataset) and customizing or “adapting” it to a more specific task or dataset.
This approach allows developers to leverage the extensive learnings of the base model while tailoring its behavior to better suit their unique needs, thereby improving the accuracy and efficiency of their specific application. This highly anticipated capability is something developers can look forward to later this year.
Chat Completed
The rise of the chat-based model used in GPT-4 is significant. Since its launch in March, the Chat Completions API now accounts for 97% of OpenAI GPT usage, effectively replacing the old free-form text prompt-based Completions API. The shift to a more structured chat-based interface has proven to be a game-changer, providing more flexibility, more specificity, and superior results.
However, these improvements do not come without a cost. OpenAI announced a deprecation plan for the old Completions API models.
Starting January 4, 2024, the old completion model will be retired and replaced with the new model as part of OpenAI's increased investment in the Chat Completions API and its efforts to optimize compute power.
"While this API will still be accessible, we will be marking it as 'legacy' in our developer documentation starting today."
Developers who wish to continue using their fine-tuned models after January 4, 2024 will need to fine-tune a replacement model on top of the new base GPT-3 model or updated models such as gpt-3.5-turbo and gpt-4.
Embedding models are deprecated
OpenAI also said that in light of these developments, the old "embedding" model will be deprecated. Users must migrate to "text-embedding-ada-002" by January 4, 2024. OpenAI has assured developers using the old model that they will bear the financial cost of re-embedding their content using these new models.
As OpenAI leads this seismic shift, it also raises questions about the future of older models and the impact on developers and companies that rely on them. This historic turn in the development of artificial intelligence highlights the rapid and relentless pace of technological innovation that is shaping the future of industries around the world.