Stop OpenAI Training Its Models On Your Chats By Turning Off History

Republished By Plato

Followers: 0

OpenAI, the Microsoft-bankrolled outfit behind the chatbot star of the moment, ChatGPT, launched a feature on Tuesday that allows users to restrict the company from using text generated in their private conversations to train large language models.

“We’ve introduced the ability to turn off chat history in ChatGPT,” OpenAI confirmed in a statement. “Conversations that are started when chat history is disabled won’t be used to train and improve our models, and won’t appear in the history sidebar.”

Users should go to settings and toggle a button to turn off chat history. OpenAI, however, will still keep records of conversations for 30 days before deleting them even if they aren’t used for training purposes. Users can also request to export their data from ChatGPT and have the file emailed to them.

Speaking of machine learning… Officials from the FTC and other US federal agencies have pledged to enforce today’s civil rights laws against AI systems that perpetuate bias.

The change comes as regulators in multiple countries investigate the free, popular chatbot over data privacy concerns. Italy’s Guarantor for the Protection of Personal Data temporarily banned the tool while it probed whether ChatGPT violated the European Union’s GDPR and its own privacy laws. Officials in Canada, France, and Spain have also launched their own investigations to scrutinize the way data is collected, stored, and used by the software.

Some large corporations like JP Morgan, Goldman Sachs, Wells Fargo, Verizon, and others have limited employees’ access to ChatGPT, fearing that their conversations could inadvertently leak sensitive information – including financial data, personal details, or trade secrets. Since language models learn to generate text based on their training data, people are worried that the tool could later produce outputs containing snippets of private information.

OpenAI also announced it is working on ChatGPT Business – a subscription tier that will allow organizations managing their own customers to have greater control over their data. Text generated in conversations by ChatGPT Business users will not be used to train OpenAI’s models as per its API’s data usage policies. ChatGPT Business is expected to be available in the next few months.

The upstart recently got some unwanted attention when a bug allowed users to read bits of conversations stored in other people’s chat histories, and see details of subscription payment plans like names and email addresses. CEO Sam Altman explained the glitch stemmed from an issue in an open source library used by ChatGPT. The website was briefly taken offline last month while the bug was fixed. ®