- Sam Altman showed Marc Benioff an unreleased OpenAI model which can clone voices, Forbes reported.
- The ChatGPT creator generated a John F. Kennedy speech in Benioff's voice over a dinner meeting.
Sam Altman created an AI voice clone of Marc Benioff giving a presidential speech, but can no longer locate the data sample, Forbes and Quartz reported.
Benioff, the Salesforce CEO, spoke at an event for analysts and the media in New York about his dinner meeting with the creator of ChatGPT, who is also his neighbor.
In his company's most recent earnings call, Benioff had alluded to Altman giving him "demonstrations of advanced technologies" but didn't then give any details.
He revealed Monday that Altman used an unpublished OpenAI model to simulate Benioff repeating the words of John F. Kennedy. Benioff's speech was reported by both Forbes and Quartz, with the latter reporting that the technology had recreated an entire speech by JFK in Benioff's voice.
Such voice clones have become more advanced in recent months, being used to create a viral AI-generated Drake song, and a "final" Beatles track.
ElevenLabs, one such popular software, says users need at least one minute of audio samples to create a custom AI voice.
Benioff said that OpenAI's model was able to accurately generate the JFK speech with just a few seconds of his voice, Forbes reported.
It turns out being the boss of one of the most exciting AI companies affords some fun parlor tricks too. Although there could also be data-protection problems.
When Benioff wanted to know where the audio sample was stored and how it was secured, Altman said there was no way to see where the data was now being stored, per Quartz.
According to Forbes, Benioff explained that while the technology is cool, it could be a concern for CEOs who need to know where their data is for compliance reasons.
AI voice cloning has also been used in a range of malicious cases, like trying to scam one woman with a $1 million ransom by pretending her daughter had been kidnapped.
And in January, ElevenLabs quickly limited access after trolls generated clips impersonating celebrities giving racist and violent speeches, Vice reported.
OpenAI did not immediately respond to Insider's request for comment, sent outside US working hours.