According to Cointelegraph: OpenAI has announced the availability of fine-tuning for GPT-3.5 Turbo, allowing AI developers to improve performance on specific tasks using dedicated data. The development has been met with both excitement and criticism from the developer community.
OpenAI explained that through fine-tuning, developers can tailor the capabilities of GPT-3.5 Turbo to their needs. For instance, a developer could fine-tune GPT-3.5 Turbo to generate custom code or efficiently summarize legal documents in German using a dataset from the client's business operations.
However, some developers have expressed concerns about the new feature. A comment from a user named Joshua Segeren suggests that while fine-tuning GPT-3.5 Turbo is interesting, it's not a complete solution. He believes that improving prompts, using vector databases for semantic searches, or switching to GPT-4 often yield better results than custom training. Additionally, setup and ongoing maintenance costs should be considered.
The base GPT-3.5 Turbo models start at a rate of $0.0004 per 1,000 tokens, while the fine-tuned versions cost $0.012 per 1,000 input tokens and $0.016 per 1,000 output tokens. An initial training fee based on data volume also applies.
The fine-tuning feature is significant for businesses and developers looking to create personalized user interactions. For example, companies can fine-tune the model to match their brand's voice, ensuring that chatbots display a consistent personality and tone that aligns with the brand identity.
To ensure responsible use of the fine-tuning feature, the training data used for fine-tuning is reviewed through OpenAI's moderation API and the GPT-4 powered moderation system. This process helps maintain the security features of the default model during the fine-tuning process and ensures that the refined output complies with OpenAI's established security standards. It also allows OpenAI to maintain some control over the data that users input into its models.
