More from Live updates from Elon Musk and Sam Altman’s court battle over the future of OpenAI

Savitt is bringing up some previously released email exchanges where Musk appeared okay with discussing what OpenAI would and wouldn’t make open source — including one where he replied “yup” to a comment that it would make sense to start being less open as AI advanced. He asks if Musk has made xAI’s own advanced versions of Grok open source — “No, but it will,” Musk says.

Musk has explained that he didn’t object to the proposed introduction of a capped profit structure initially at OpenAI (and also didn’t review it very closely), and Savitt is asking if he knew what the cap was for Microsoft’s investments in the company — Musk doesn’t seem clear on it. Savitt asks whether Musk had a lawyer set terms and conditions for his donations. Musk answers: “No, but it was obviously started as a nonprofit, and in the founding charter it says it will not be to the financial benefit of any person?” The apparent intended gist is that Musk didn’t set clear terms he can point to OpenAI or Altman violating.

Elon Musk is on the stand, to continue the cross-examination from yesterday. If you’ve read Musk depositions or heard previous crosses, this kind of arguing and filibustering is pretty standard behavior. But I think this is the jury’s first encounter with it, and it’s hard to know how they’re going to take it.

We are still dealing with the pretrial motions about the boundaries on safety questions.

We are having an argument about which expert issues are going to be allowed. “We aren’t going to get into issues of catastrophe or extinction,” YGR says. Musk’s lawyers are not happy about this: “We all could die as the result of artificial intelligence.”

YGR has just sat down on the bench. Jury’s not here yet, so we are dealing with some motions and issues.

I’ve just left the court after one of the most tiresome cross-examinations I have ever had the misfortune to witness. Elon Musk really wants us to think he’s just a poor simple country CEO, who is being maliciously tricked by a big-city lawyer. I’m not buying it and I don’t think the jury is, either

Well there goes all my fun! I was so excited to hear about Mechahilter in open court — and, in addition, Sam Altman apologizing to a Canadian town for OpenAI not flagging a mass shooter. But since Musk did bring up the difference in for-profit and not-for profit motives, there is a limited line of questioning that can proceed.

There are a few matters the lawyers want to discuss with YGR. Musk’s testimony may have opened the door to questions about xAI and its safety record. I do love hearing the phrase “opened the door.”

Savitt hasn’t significantly raised the issue yet, but he’s started hinting at the idea that Musk’s safety-last approach at xAI might undercut his credibility. He asks Musk if it’s important to instill good values in AI systems, and after Musk agrees, asks if racist or sexist training materials could have a negative impact. Musk says systems wouldn’t necessarily absorb those values, and Savitt follows up to mention AI discrimination — and more specifically, an anti-algorithmic discrimination law in Colorado that Musk and xAI have been fighting against.

Savitt briefly explores a line of questioning about Grok, but it ends quickly. (It could be brought up for discussion later.) He moves on to asking about Musk’s purported commitment to care deeply about AI safety. Has Musk ever posted on Twitter/X about AI regulation? Musk doesn’t know off the top of his head. Savitt also sounds dubious of the idea that Musk spoke with Obama about AI safety, asking if there are any press reports or statements from the White House. Then, Savitt asks if Musk has spoken with the current president about the issue, especially since the former White House AI czar, David Sacks, is one of Musk’s fellow Paypal Mafia members.

Elon Musk is trying to outlawyer the lawyer on cross. At one point YGR has to intervene to get him to answer a yes or no question with “yes.” At another, he raised his voice to lecture Savett — not sure if the jury noticed the, uh, well it wasn’t quite yelling but it came pretty close.

Savitt asks Musk about a term sheet for OpenAI’s for-profit shift, which Sam Altman sent Musk to examine in 2018, and Musk admits “I didn’t read the whole document.” (It’s apparently four pages long.) The attorney then brings up a deposition in which Musk says multiple times that he doesn’t think he read it or looked closely at it. Musk gets testy and raises his voice as he’s asked to identify any communications where he objects to the proposed changes in 2018 or 2019. He repeats, once more, that he’s okay with a nonprofit having a capped for-profit arm — which Savitt says isn’t his question.

According to messages between Musk and Zilis, Musk thought OpenAI stood essentially zero chance of succeeding against Google Deepmind, particularly if he focused on AI at Tesla. Zilis remained close with OpenAI and apparently offered Musk updates, including on its for-profit plans. Savitt mentions, again, that Musk didn’t object to information about this for-profit shift at the time — and again, Musk says he didn’t have an issue with a capped for-profit structure that would flow into a nonprofit.

He calls it one of several companies working toward the goal and likely not the one that will win the race. Savitt goes back to Tesla briefly, noting that Musk apparently hoped to build it into an AI powerhouse at one point. He displays an email Musk sent to Gabe Newell about OpenAI, saying he “lost confidence” that OpenAI would serve as an effective counterweight to Google Deepmind and decided to attempt that through Tesla instead.

After a quick breather, Musk is on the stand again — now the topic is SpaceX, which xAI has been rolled into. The company is preparing for an IPO later this year. Musk says he can’t answer questions about SpaceX because of this.

Altman’s lawyer mentions Musk building an “AI-enabled robot army” while talking about the proposal to merge Tesla and OpenAI. Musk jumps in, saying he wants to make clear that while he has in fact referred to running a “robot army” at the company, he did not mean the term “robot army” in a “military sense.” Glad we’ve cleared that up.

Musk is not terribly happy with Savitt, who continues to run through the history of entanglements between Musk’s own companies and OpenAI. After telling Musk that “I’m trying to put the questions as fairly as I can, I’m doing my best,” (“That is not true,” Musk responds) Savitt brings up a deposition where Musk said he’d discussed a merger between Tesla and OpenAI. In email exhibits, Karpathy (who’d been hired away to Tesla by then) suggests OpenAI is burning cash and its funding structure won’t let it compete with Google, saying the best path forward might be “a for profit pivot” and attaching itself to Tesla. Any other big tech player, Karpathy said, could suffer from “incompatible company DNA.”

Musk previously mentioned that he hired researcher Andrej Karpathy at Tesla when he left OpenAI. Savitt asks whether, as a member of the OpenAI board at the time, he had a responsibility to suggest Karpathy stay at OpenAI and avoid poaching. “I think people should have a right to work where they want to work,“ Musk responds. A 2017 email suggests Musk knew how important Karpathy was:

Musk quibbles his way through a line of questioning from Savitt, who asks if he was aware that cutting off most funding to OpenAI in 2017 would create financial pressure as OpenAI sought to get more compute. Musk repeats multiple times that he was losing faith in OpenAI, he was concerned they were going in the wrong direction, and that’s why he ended his donations. Savitt and Musk then spar over how much “sweat equity” Musk accrued at OpenAI — Savitt notes, for instance, he didn’t write the code for a consequential project where OpenAI’s bots learned to beat humans at DOTA 2. Musk retorts that he called Satya Nadella to get compute for the project and suggested focusing on it. “I got along with almost everyone almost all the time,” he adds.

Savitt is asking Musk about the equity discussions — the ones where Musk wanted initial control of OpenAI. Musk acknowledges his plan would have given him unilateral decision-making power at first but says that’s “standard practice.” He reiterates that it would have changed quickly as more investors joined up and insisted on board seats; over time, he says, he’d have had to relinquish control.

Musk is still fending off questions about various messages where he’s mentioned the possibility of a for-profit component at OpenAI, like an email to associates at Neuralink in 2016:

In an email a month before OpenAI was incorporated, Musk apparently wrote to Altman suggesting that “a standard c-corp with a parallel nonprofit” could work for the entity. Questioning about this quickly runs off the rails as Musk starts accusing Savitt of trying to “trick” him, comparing his questions to asking Musk if he’s stopped beating his wife. Judge Gonzalez Rogers is having none of this and cuts him off.

“Your questions are not simple, they are designed to trick me, essentially. If you ask a question where there is no possible simple answer, I must give a longer answer.”