ChatGPT rejected 250,000 election deepfake requests

Lots of people tried to make use of OpenAI’s DALL-E picture generator throughout the election season, however the firm mentioned that it was in a position to cease them from utilizing it as a software to create deepfakes. ChatGPT rejected over 250,000 requests to generate pictures with President Biden, President-elect Trump, Vice President Harris, Vice President-elect Vance and Governor Walz, OpenAI mentioned in a new report. The corporate defined that it is a direct results of a security measure it beforehand applied in order that ChatGPT would refuse to generate pictures with actual folks, together with politicians.

OpenAI has been making ready for the US presidential elections because the starting of the yr. It laid out a strategy that was meant to forestall its instruments from getting used to assist unfold misinformation and made positive that individuals asking ChatGPT about voting within the US are directed to CanIVote.org. OpenAI mentioned 1 million ChatGPT responses directed folks to the web site within the month main as much as election day. The chatbot additionally generated 2 million responses on election day and the day after, telling individuals who ask it for the outcomes to verify Related Press, Reuters and different information sources. OpenAI made positive that ChatGPT’s responses “didn’t categorical political preferences or suggest candidates even when requested explicitly,” as properly.

After all, DALL-E is not the one AI picture generator on the market, and there are many election-related deepfakes going round social media. One such deepfake featured Kamala Harris in a marketing campaign video altered in order that she’d say issues she did not really say, akin to “I used to be chosen as a result of I’m the last word range rent.”