Loading content
Preparing Central Tech…
Loading content
Preparing Central Tech…
Loading content
Preparing Central Tech…
Adi Robertson has been covering the intersection of technology, culture, and policy at The Verge since 2011. Her work includes writing about DIY biohacking, virtual and augmented reality, copyright law, online free expression, and the history of computing. You have probably seen her in a VR headset.
She says the starting point was Sutskever reaching out to have a conversation where he expressed serious concerns about Altman. It was a “pattern of behavior” that included issues with “honesty and candor” that led to the firing, not any one action. Toner has already laid out some of this in a 2024 podcast, and it’s similar to Murati’s testimony.
And he thinks posts are still called tweets. (The post is about AI creating so much wealth that “everyone can have a penthouse if they want.”) We’re getting some meandering analysis of AI creating prosperity, but even the judge seems frustrated at this point.
Russell is asked next about risk. He starts with individual risk, including algorithmic discrimination, which he says is “much more widespread” than people expect. He also mentions the possibility of AI systems reinforcing “delusional beliefs” — like, in his example, “Vladimir Putin is an alien.” Then there’s systemic risk — like AI systems taking jobs at a large scale. Russell mentions computer science students having trouble getting work already, and he expects it will accelerate.
As Russell keeps going, the opposition lawyer objects that this is speculative, and Judge Gonzalez Rogers asks to move on — which Musk’s lawyer does by promptly asking again about jobs. Another objection is sustained, and Russell starts talking about people using AI to magnify harm, like cyberattacks. He mentions Mythos, and the opposition lawyer gets it stricken from the record as outside the scope. None of this particularly relates to OpenAI so far — it’s basically a generic overview of AI risks, which Musk has been focusing on as central to the trial, despite some limits from YGR. And we’ve already run through 20 minutes of the 30 allotted for direct testimony.
Russell is first asked about the positive aspects of AI — he mentions medical technology and AlphaFold, which can predict protein structures. Also mentions Waymo self-driving cars, which he describes as “extremely safe.” There can also be “broad economic benefits” where systems can be used to develop software, “where the AI can actually write software much faster and cheaper than human beings.”
Judge Yvonne Gonzalez Rogers says she got some “May the 4th be with you” jokes from her kids today as we get started. Apparently she appreciated them. After that, the day kicks off with expert testimony from computer scientist Stuart Russell.
Emails going as far back as 2015 give a glimpse into the foundations of OpenAI and the early tensions at the company.
We’re looking at a summary of about 60 donations to OpenAI, which Birchall says were directed by Musk, with Birchall helping execute all of them.
Musk’s back from break, reiterating that he had reason for waiting as long as he did to file suit against OpenAI — and saying his initial understanding of OpenAI’s agreement with Microsoft was that it didn’t violate the mission of the charity. “I don’t think I had a basis for filing a lawsuit before I did,” Musk says. He also refers to xAI as the smallest of the AI players, coming after Anthropic, OpenAI, Google, and Chinese AI models.
Apparently he wasn’t 100 percent confident in yesterday’s clarification, because Molo asks Musk to clarify whether the “AI-enabled robot army” mentioned in cross-examination is a military army. “No, we do not make any weapons,” Musk says. The point of his using the term was that “if we made a lot of robots we need to make sure they’re safe and don’t turn into a Terminator situation … You see, in the movie, it’s not a good situation.”
Judge Gonzalez Rogers asks Musk to sum up the plot of Terminator in one sentence. “Worst-case situation is AI kills us all, I suppose,” he says.
With that, the jury leaves for a break.
Under questioning from Molo, his own lawyer, Musk tries to establish that he wasn’t causing harm to OpenAI. He says that as far as he knows, OpenAI wasn’t unable to cover any critical expenses because he ended his donations. He didn’t ask Andrej Karpathy to leave and join Tesla, only hired him after he said he was leaving OpenAI. Neuralink (while it was authorized to do so apparently) didn’t poach anyone from OpenAI as far as he knows. Did he seriously recruit anyone from OpenAI for Tesla besides Karpathy? “I don’t think so.” He reiterates that Tesla isn’t currently working on AGI, despite a recent tweet indicating it would achieve it.
Musk also repeats that he “did not read the fine print” on the term sheet for OpenAI’s for-profit wing. Molo brings up an email from Altman (forwarded to Musk by Zilis) about the draft that reads: “We did this in a way where all investors are clear they should never expect a profit, see purple box below.” On the stand, Musk says “I assumed he meant what he said.”