ALBAWABA - OpenAI, ChatGPT maker and key player in the artificial intelligence industry, has revealed that its most sophisticated model, GPT-4o, may now be fine-tuned with custom training data and adapted by developers for use in commercial applications.
Developers who are working on OpenAI's application programming interface (API) may now utilize their own datasets to enrich GPT-4o's knowledge base with confidential information and influence how the model reacts to certain inquiries, according to a blog post published by the company.
As an example, a firm that manufactures eyeglasses may modify an artificial intelligence model in order to make it suitable for use as a customer service chatbot that is able to respond to inquiries about frames as well as the particulars and distinctions between different kinds of lenses.
Up until this point, neither the GPT-4o nor its predecessor, the GPT-4, had the capability to be fine-tuned, according to Bloomberg. On the other hand, OpenAI has made it possible for consumers to fine-tune a number of its various models, such as the GPT-4o mini, which is a more reasonably priced and simplified version of the GPT-4o.
For each million tokens used to fine-tune GPT-4o, customers must pay $25. The more data required to train the model on, the more tokens the model will need to analyze. Tokens are bits of data that are transformed into a format that AI models can understand, according to Inc. To put things in perspective, OpenAI estimates that one million tokens is about the same as 2,500 pages in a typical book.
In its blogpost, OpenAI touches on privacy and confidentiality concerns for business wishing to fine tune an AI model for their needs, saying “Fine-tuned models remain entirely under your control, with full ownership of your business data, including all inputs and outputs. This ensures your data is never shared or used to train other models.”
