OpenAI Enables GPT-3.5 Turbo to be Customized
OpenAI, the Microsoft-backed company behind the creation of ChatGPT, has announced that users of its GPT-3.5 Turbo AI model can now fine-tune and customize the model according to their preferences and use cases. This change allows companies to effectively adopt these models at a larger scale and thus adapt to their specific needs.
Additionally, the company has revealed that a similar feature will be rolled out to GPT-4 later this fall.
According to early tests by OpenAI, a “fine-tuned” version of GPT-3.5 Turbo has the potential to match or even exceed “basic GPT-4-level capabilities for certain narrow tasks.” And the data that a given company uses to fine-tune the model for its own application — not shared with OpenAI — ensures both security and privacy.
OpenAI has outlined several applications and immediate benefits of AI model fine-tuning:
Custom tone
By fine-tuning a template, companies can ensure that it aligns with their specific brand “tone,” making it more appropriate for their brand identity. “A company with a recognizable brand voice can use fine-tuning to make the design more consistent with their tone,” OpenAI said.
Improved maneuverability
Simply put, with the help of optimization and fine-tuning, companies can modify the model to better suit their own operations. For example, the model can be fine-tuned to respond in German or other languages if necessary, which can lead to better efficiency.
Consistent output formatting
Another immediate benefit of fine-tuning is an improvement in the model’s ability to formulate responses consistently. OpenAI notes that this is “an important consideration for applications that require a specific response format, such as code completion or writing API calls,” and that a developer can use the fine-tuning “to more reliably convert user prompts into high-quality JSON snippets that can be used with their own systems.”
In addition, OpenAI has listed additional benefits, including the ability to handle up to 4k tokens – double that of previous fine-tuned models. Businesses can even reduce prompt size by up to 90%.
