GPT-3.5 can now adopt the corporate identity of your company. OpenAI is launching a new API that allows companies to refine the AI ​​model.

OpenAI expands its range of APIs for ChatGPT. Paying customers can now develop a customized version of GPT-3.5 that better suits their needs. At the moment this is possible for the “turbo version” of GPT-3.5, but from autumn OpenAI will also offer this option for GPT-4.

Companies that sign up for the API start with the base version, which will be trained with data until September 2021. They can systematically fully customize the chatbot by feeding it internal data and programming specific commands in the system. Setting ChatGPT to always reply in Dutch in a tone that fits your brand’s style or converting reply prompts to JSON files are just some of the features OpenAI lists in the announcement blog.

According to OpenAI, the big benefit of fine-tuning GPT-3.5 is that the output is much better tailored to your specific business needs. The functions of a fine-tuned version of GPT-3.5 correspond to those of GPT-4 for certain commands. Prompts can also be shortened by up to ninety percent.

keep a secret

To address privacy concerns, the data you feed into the system via the API also stays internal and is not shared with other users. The launch of ChatGPT itself should convince companies that don’t want employees to work with the chatbot. Recent research shows that many companies are banning or considering banning ChatGPT for security reasons.

Of course, you pay a price for using the API. In the training phase, you pay $0.008 per thousand tokens, the usage cost is $0.012-0.016 per thousand tokens. OpenAI can put all the revenue generated by the GPT models to good use.

What OpenAI does is also very similar to what Microsoft wants to offer with Bing Chat Enterprise. This is also a version of the chatbot that is geared towards business use and keeps the data internal. OpenAI is again in full competition with its main sponsor.