OpenAI Collaborates with Scale AI to Enable Businesses to Fine-Tune GPT-3.5 for Enhanced Performance

OpenAI, the pioneering AI startup known for the widely popular chatbot ChatGPT, is set to collaborate with third-party vendors to simplify the process for enterprises to fine-tune its AI models using their own data.

In a recent announcement, OpenAI revealed that it will partner with Scale AI, a San Francisco-based data labeling startup, to integrate Scale AI's fine-tuning tools with OpenAI's GPT-3.5 text-generating model. (GPT-3.5, which precedes OpenAI's flagship model GPT-4, offers the capability to comprehend both text and images.)

Fine-tuning allows developers to customize an AI model for specific applications. For instance, a company could adapt the model to reflect its unique brand voice, or program it to handle inquiries in a specific language. Scale's clients will be able to prepare and enhance their data using Scale's Data Engine platform, enabling them to fine-tune GPT-3.5 with this data and incorporate features such as referencing or citing proprietary data in responses.

Scale has stated that fine-tuned GPT-3.5 models will undergo review by human experts to ensure they surpass performance benchmarks and adhere to safety standards.

“The collaboration merges OpenAI’s GPT-3.5 fine-tuning APIs with Scale’s extensive experience in fine-tuning and data management, developed over more than seven years of delivering innovative AI solutions,” said Scale CEO Alexandr Wang in an email interview. “Scale has become OpenAI’s preferred fine-tuning partner to launch highly effective models, and we anticipate deepening our partnership. OpenAI has selected us for our expertise and proven ability to deliver impactful, high-quality models for our clients.”

This news follows OpenAI's announcement that it will allow customers to fine-tune GPT-3.5 Turbo, a lighter iteration of GPT-3.5, using its fine-tuning API. The announcement also hinted at the development of an in-house tuning tool, although it's uncertain if this tool is intended to complement platforms like Scale AI or serve a different function altogether.

Regardless, OpenAI COO Brad Lightcap emphasized the collaboration with Scale AI as a way to provide companies that tailor GPT-3.5 with additional services and expertise that go beyond OpenAI's offerings. “We aim to support our clients wherever they are. We're thrilled to partner with industry leaders like Scale, which helps us meet customer needs as they implement AI into their operations,” he said via email. “This new partnership enables us to extend our capabilities to a wider range of companies, leveraging Scale’s leadership and experience in helping organizations effectively utilize enterprise data for AI. Our clients can now fine-tune OpenAI models through us while also reaping the benefits of Scale’s enterprise AI expertise and Data Engine.”

When asked whether new fine-tuning partners for OpenAI's models, such as GPT-4—which is expected to gain fine-tuning capabilities this fall—might emerge in the future, Lightcap did not provide any specific information.