OpenAI makes GPT-3.5 Turbo available for custom tuning

Businesses may now fine-tune GPT-3.5 Turbo using their own data, according to OpenAI.

According to OpenAI, the generated custom model can match or outperform GPT-4 in some tasks. The business says it will offer up the arguably more advanced GPT-4 for the same purpose later this fall.

Fine-tuning essentially allows firms to hone ChatGPT into a more concentrated model that is exceptionally efficient for certain tasks. The supervised training would create a bot that is customized to the client organization, such as providing dependable responses in a certain language or using more succinct wording. Previously, commercial clients were limited to GPT-3 models such as Davinci-002 or Babbage-002.

The model, like GPT-4, would be pre-trained until September 2021 before being fed company data. According to OpenAI, no data, input, or output will be used to train models outside of the client organization.

Other applications include training the bot to replicate brand tones so they are more consistent — think ad content or internal communications authored in part by AI (not that we don't already see plenty of that). It could be used by software businesses for routine programming, such as API calls, or to consistently format and complete snippets of code.

The GPT-3.5 Turbo is a model family introduced by the company earlier this year that it claims is appropriate for non-chat use cases. It can comprehend 4,000 tokens at a time, which OpenAI claims is quadruple what prior models could. Early testers were able to produce 90 percent shorter prompts after priming GPT-3.5 with fine-tuned instructions, according to the business.