![](https://kernel.news/wp-content/uploads/2023/03/ChatGPT-Robot-780x470.jpeg)
OpenAI has made a significant announcement that businesses can now customize GPT-3.5 Turbo using their proprietary data. According to OpenAI, the resulting tailored model can potentially match or even surpass GPT-4 in performance for specific tasks. Furthermore, OpenAI plans to extend this fine-tuning capability to the arguably more advanced GPT-4 later this fall.
Fine-tuning allows businesses to refine ChatGPT into a model that is particularly adept at certain tasks. This supervised training results in a bot tailored to the client company, ensuring it provides reliable responses in a specific language or with more precise wording. Previously, businesses could only fine-tune GPT-3 variants, such as davinci-002 or babbage-002.
The model will be pre-trained, similar to GPT-4, up to September 2021, after which it will be trained with company-specific data. OpenAI has assured that none of this data, including any input or output, will be utilized to train models outside of the client’s company.
One of the potential applications of this fine-tuning is to train the bot to mimic brand voices, ensuring consistency in communications, whether it’s advertising copy or internal communications. Software companies could also utilize it for routine coding tasks, such as API calls or to reliably format and complete code snippets.
GPT-3.5 Turbo, introduced earlier this year by OpenAI, is designed for a broader range of use cases beyond just chat. It can process up to 4,000 tokens at once, which is double the capacity of previous models. Early testers have reported that after priming GPT-3.5 with fine-tuned instructions, they could create 90% shorter prompts. The pricing for GPT-3.5 is set at $0.0080 per 1,000 tokens for training, and $0.0120 per 1,000 tokens for both input and output usage.
In related news, Microsoft is also offering customizable GPT-based models through its AI Builder and Power Virtual Agents services. These are designed to be linked to a company’s internal data to craft responses, summarizing information or generating content for email campaigns. Similar to OpenAI’s models, Microsoft’s AI bots can be connected to company data to produce responses based on the business’s knowledge base.