OpenAI allows for customized tweaking of the GPT-3.5 Turbo

Businesses may now fine-tune GPT-3.5 Turbo using their own data, according to a recent announcement from OpenAI. According to OpenAI, the resulting bespoke model can match or even outperform GPT-4 for some jobs. The business claims that it will make the arguably more sophisticated GPT-4 available for the same purpose later this fall.

Businesses can effectively fine-tune ChatGPT to a more concentrated model that is especially effective for specific tasks. The supervised training would create a bot that is tailored to the client business and would provide, for example, dependable responses in a particular language or with shorter words. Business clients had to use GPT-3 variations like the davinci-002 or babbage-002 up until this point.

Similar to GPT-4, the model would be pre-trained up until September 2021 before being fed business data. None of the data, along with any inputs or outputs, will be utilized to train models outside of the client organization, according to OpenAI.

Other applications include making sure the bot is taught to replicate brand tones so they are more consistent — think of ad content or internal communications that are at least partially produced by AI (not that we don’t already see plenty of this). It could be used by software businesses for routine code, such as API calls, or to consistently format and finish pieces of code.

The GPT-3.5 Turbo model family, which the firm introduced earlier this year, is appropriate for use cases that aren’t chat-specific, according to the company. It can process 4,000 tokens at once, which according to OpenAI is twice as many as the models that were previously available. The business also stated that after priming GPT-3.5 with refined instructions, early testers were able to create prompts that were 90% shorter.

The cost of GPT-3.5 is $0.0080 for training, $0.0120 for input usage, and $0.0120 for output from the chatbot per 1,000 tokens.

A company’s internal data can be integrated to Microsoft’s AI Builder and Power Virtual Agents services to create replies using refineable GPT-based models. They are promoted by Microsoft as a tool to compile data or create content for email campaigns. Microsoft’s adaptable AI bots can link to corporate data to produce responses from a company’s knowledge base, just like OpenAI’s fine-tuning bots can.