During the US election season, many tried to create deepfakes using OpenAI's 'Dal-e' image generator. However, ChatGPT has prevented the creation of more than 250,000 such deepfakes.
According to a new report by OpenAI, this time ChatGPT was directed to create fake photos or deepfakes of President Biden, Donald Trump, Vice President Kamala Harris, James David Vance and Governor Walz. However, the chatbot is able to reject these requests. OpenAI already adds a security mechanism, whereby chatgpt will refuse to create images of real people, especially politicians.
OpenAI has been preparing for the US presidential election since the beginning of this year. The company has adopted various strategies to ensure that its tools are not used to spread false information. This strategy ensures that those interested in voting on ChatGPT are directed to the CanIVote.org website for information.
OpenAI said it generated 1 million or 1 million replies to ChatGPT users on the website a month before the election. The chatbot generated 2 million or 2 million responses on and after the US election day. Through these answers, people who want to know the results of the vote are sent to the Associated Press, Reuters and other media. OpenAI also confirmed that ChatGPT's answers did not express 'political bias' or recommend any candidate for voting. Even if such questions are asked clearly.
Dal is not the only AI image generator and many election-related deepfakes are also spreading on social media. This includes a DeepFolk video of Kamala Harris. Where Kamala makes Harris say things he never said.
For example, the video says 'I was selected because I am the highest diversity hire.' Such fake videos create confusion and are used to spread false information.
