ChatGPT rejected 250,000 election deepfake requests

Lots of people tried to make use of OpenAI’s DALL-E picture generator in the course of the election season, however the firm mentioned that it was capable of cease them from utilizing it as a device 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 carried out in order that ChatGPT would refuse to generate pictures with actual individuals, together with politicians.

OpenAI has been getting ready for the US presidential elections for the reason that starting of the 12 months. It laid out a strategy that was meant to stop its instruments from getting used to assist unfold misinformation and made positive that folks asking ChatGPT about voting within the US are directed to CanIVote.org. OpenAI mentioned 1 million ChatGPT responses directed individuals 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 truly say, comparable to “I used to be chosen as a result of I’m the final word variety rent.”