“Clearly our customers don’t want us to train our model on their data, so we changed our plans”, said Sam Altman, Friday, May 5, to the American television channel CNBC. The CEO of OpenAI, the California-based start-up that develops the famous ChatGPT virtual chatbot, has announced that this important change has in fact taken place “a while ago” : since March 1, it seems.
Capable of generating tailor-made answers to any question after being trained on huge volumes of textual data, ChatGPT has aroused fascination and concern since its launch in November. The protection of the personal data of its users, in particular, does not appear to be guaranteed by this tool, which operates particularly opaquely. The company refuses, for example, to reveal the data sets used for its training.
In any case, since March 1, the information provided by the user at the time of his virtual dialogue with the chatbot is no longer memorized to improve the tool. With one detail: this protection only concerns companies with a paid subscription. This allows them to use OpenAI technology to build their own software solutions.
Internet users using the public version of ChatGPT continue to see their virtual conversations with the chatbot exploited in order to improve it.
Samsung and Amazon
This announcement from the CEO of OpenAI therefore aims above all to reassure companies. Some have seen, in recent weeks, information from their internal documentation “emerge” during requests made outside. On May 2, the Samsung Electronics group declared that it had prohibited some of its employees from using such tools after noting in an internal document « cas d’utilisation abusive ». Goldman Sachs had previously made a similar announcement.
As early as January, Amazon instructed its employees not to share confidential information with ChatGPT, lest it appear in the chatbot’s responses. Some technicians in the company had started using ChatGPT to help them write computer code.
“This raises issues of personal data, but also copyright and even business or medical secrecy”, explains Arnaud Latil, lecturer in law at Sorbonne University. But for him, the question of the legality of ChatGPT arises above all for training data. The billions of Internet pages that the tool had to ingest to “learn” to generate texts undoubtedly contain a plethora of personal information.
Again authorized in Italy
The Italian Personal Data Protection Authority blocked ChatGPT the whole month of April, accusing it of not respecting the General Data Protection Regulation (GDPR) and of not having a system to verify the age of minor users.
On April 28, OpenAI announced that it was functional again in Italy. The company assured that it will now publish on its site information on how it “collection” et “uses training-related data”. A ” greater visibility » will be given to the personal data policy on the home page of ChatGPT and OpenAI.
When the user connects for the first time to ChatGPT, he is informed that the data provided in the conversations will be exploited, but he does not have the means to oppose it. Consent is one of the legal bases provided by the GDPR to authorize such exploitation.
#ChatGPT #reassure #businesses #privacy #policy