OpenAI hasn’t trained its AI large-language models such as GPT with paying customer data “for a while,” CEO Sam Altman told CNBC on Friday.
“Customers clearly want us not to train on their data, so we’ve changed our plans: We will not do that,” Altman told CNBC’s Andrew Ross Sorkin.
OpenAI’s terms of service were quietly updated March 1, records from the Internet Archive’s Wayback Machine show. “We don’t train on any API data at all, we haven’t for a while,” Altman told CNBC. APIs, or application programming interfaces, are frameworks that allow customers to plug directly into OpenAI’s software.
OpenAI’s business customers, which include Microsoft, Salesforce and Snapchat, are more likely to take advantage of OpenAI’s API capabilities.
But OpenAI’s new privacy and data protection extends only to customers who use the company’s API services. “We may use Content from Services other than our API,” the company’s updated Terms of Use note. That could include, for example, text that employees enter into the wildly popular chatbot ChatGPT. Amazon reportedly recently warned employees not to share confidential information with ChatGPT for fear that it might show up in answers.
The change comes as industries grapple with the prospect of large-language models replacing material that humans create.
The Writers Guild of America, for example, began striking Tuesday after negotiations between the Guild and movie studios broke down. The Guild had been pushing for limitations on the use of OpenAI’s ChatGPT for script generating or rewriting.
Executives are equally concerned about the impact of ChatGPT and similar programs on their intellectual property. Entertainment mogul and IAC chairman Barry Diller has suggested that media companies could take their issues to the courts and potentially sue AI companies over the use of the creative content.
