Elon Musk Accuses OpenAI Of “Straight Up” Lying About Training A.I.

This is not investment advice. The author has no position in any of the stocks mentioned. Wccftech.com has a disclosure and ethics policy.

SpaceX and Tesla chief Elon Musk, in a talk given at the New York Times’ Digital Events Summit, flat-out refused claims by A.I. companies that they do not train on copyrighted data. Musk’s talk was widely quoted by the press for his expletive directives to advertisers to stop using X if they feel like it after the billionaire was embroiled in a controversy resulting from political frictions in the aftermath of the Israel-Palestine conflict. The talk saw Musk share his thoughts about the ouster of OpenAI’s current CEO, Sam Altman, and OpenAI’s former head of research, Ilya Sutskever.

OpenAI Should Be “Super Closed Source For Maximum Profit AI,” Says Musk

During the talk, Musk shared some of the reasons behind the founding of OpenAI. He shared that one of the biggest reasons that the company was set up was because most of A.I. talent was with Google’s DeepMind. He remained critical of OpenAI becoming a for-profit company – a shift that is believed to be behind Altman’s blockbuster ouster and return from the company earlier this month.

Musk revealed that he has been interested in A.I. since his college days, but he did not engage in his interests because he was uncertain whether the benefits or the drawbacks of the groundbreaking new technology would be greater. In contrast, according to him, making life multi-planetary or investing in sustainable energy are guaranteed to generate good outcomes from the very beginning.

He highlighted the early days of OpenAI and shared that it was meant to be open source. However, now, Musk believes that OpenAI should be named “Super Closed Source For Maximum Profit AI”, in an illustration of a rather ironic outcome that stands in sharp contrast to OpenAI’s foundational principles.

“What I care about is the reality of goodness, not the perception of it. And what I see all over the place is people who care about looking good, while doing evil” stated Musk during his talk at the New York Times’ DealBook Summit with Andrew Ross Sorkin. Image: New York Times Events/YouTube

Musk wondered what could stimulate such a fundamental shift as he commented:

It’s gone from an open source, foundation, a 501(c)(3), to. . .suddenly it’s like a $90 billion full profit corporation with closed source. So, I don’t know how you go from here to there. But it seems like, I don’t know, how you get. I don’t know, is this legal? I’m like, ‘is this legal?’

Musk also believes that the fact that Sutskever “felt strongly enough to want to, you know fire Sam, well, I think the world should know what was that reason?” He also tried to investigate the reasons behind Altman’s ouster, but Sutskever remained unreachable, and others were unhelpful. Musk was the one who brought Sutskever to OpenAI, and the decision proved to be an irreparable rift between him and Google’s founder, Larry Page.

The conversation then shifted to training A.I. models, and it saw some of Musk’s sharpest comments during his talk. An A.I. model feeds on copious amounts of data to train its algorithms, which enables it to reach new conclusions on its own. Access to large data sets makes A.I. companies stand out from their peers, and while training on copyrighted data is controversial, Musk believes that all A.I. companies – including OpenAI – use such data.

Data is probably more valuable than gold,” and Musk is optimistic about the potential that X (formerly Twitter) offers when it comes to training A.I. models.

When it comes to not using copyrighted data, Musk revealed:

Yeah, that’s a huge lie.. .  These A.I. . .well these A.I.s are all trained on copyrighted data, obviously.

The host, Andrew Ross Sorkin, further probed Musk and asked if all A.I. companies, including OpenAI, were lying when they stated that they do not train on copyrighted data.

In response,  Musk remained steadfast:

Yeah, that’s a lie. Yeah, it’s a straight up lie. 100%. Obviously it’s been trained on copyrighted data.

Share this story