OpenAI is making impressive strides, revealing successive enhancements that empower users to tweak two of its most talked-about AI models.
Just last month, the company announced that ChatGPT users can now provide "custom instructions" to personalize the chatbot's responses. OpenAI explained that fine-tuning is an option for GPT-3.5 Turbo, allowing AI developers to achieve even better performance on specialized tasks using specific data.
These advancements might help the company maintain its edge over well-funded competitors like Google's Bard or Anthropic's Claude.
"We've recently introduced fine-tuning for GPT-3.5 Turbo," OpenAI shared on Twitter. "This fine-tuning lets you train the model using your company's data and deploy it extensively."
"Initial tests have shown that fine-tuned GPT-3.5 Turbo can match or even surpass GPT-4 on specific tasks," the announcement highlighted.
We've just launched fine-tuning for GPT-3.5 Turbo! Fine-tuning lets you train the model on your company's data and run it at scale. Early tests have shown that fine-tuned GPT-3.5 Turbo can match or exceed GPT-4 on narrow tasks: https://t.co/VaageW9Kaw pic.twitter.com/nndOyxS2xs
— OpenAI (@OpenAI) August 22, 2023
OpenAI showed that through fine-tuning, developers can directly customize the abilities of GPT-3.5 Turbo to suit their needs. For example, a developer might fine-tune GPT-3.5 Turbo to create specific code or accurately summarize legal documents in impeccable German, using existing data from their organization.
This feature is incredibly valuable for organizations and developers working on tailored user experiences. For instance, companies can fine-tune the model to align with their brand's voice, ensuring that a chatbot has a compatible personality and tone.
The potential of customization is also evident in the Stable Diffusion development community. Fine-tuned SD v1.5 models have achieved a level of quality that surpasses the base model, the more capable v2.1, and can even be favorably compared to the top-tier SDXL that was recently introduced.
Moreover, the benefits of fine-tuning extend to better control, consistent output formatting, reduced prompt sizes, and faster API responses with fewer costs, according to OpenAI. For instance, prompt lengths could decrease by up to 90%, speeding up workflows and cutting expenses.
While Basic GPT-3.5 Turbo models start at $0.0004 per 1,000 tokens (the fundamental unit of information processed by a Large Language Model), fine-tuned versions are pricier at $0.012 per 1,000 input tokens and $0.016 per 1,000 output tokens. The initial training phase also incurs costs based on the data size. However, the added expense might be justified considering the customization potential.
This is in addition to the "custom instructions" introduced for ChatGPT Plus users in July. For instance, users can specify their preferred coding language to ensure ChatGPT consistently suggests Python solutions. OpenAI also recommends other customization options such as location, hobbies, aspirations, and preferred tone.
Custom instructions enable users to shape ChatGPT into a personalized digital assistant tailored to their specific needs. Every interaction will adhere to these guidelines, reducing the need to repeat preferences. Instead of an entirely new model, this upgrade represents a model with a different "mindset," so to speak.
OpenAI has implemented safeguards to ensure responsible use of the fine-tuning tool. "To maintain the safety features of the default model during the fine-tuning process, the fine-tuned training data undergoes our Moderation API and a GPT-4 powered moderation system," explains OpenAI. This approach aims to identify and mitigate potentially risky training data, ensuring that even customized output aligns with OpenAI's safety standards.
This also implies that OpenAI retains a certain level of control over the data users input into their models.
Between fine-tuning and tailored instructions, OpenAI is providing customers with more flexibility to shape models according to their precise requirements. As the competition for dominance in generative AI continues, customization could be the next frontier giving OpenAI an advantage. However, for now, these capabilities remain primarily available to paying clients.
