Christina Criddle

Tämä on ensimmäinen kerta, jossa olen kuullut hänet puhumaan tällä kertaa. Se oli todennäköisesti hetki, ja periaatteessa tutkimuskeskukset ovat vastaanottaneet, että hän oli yllättänyt donoittamaan noin 38 miljoonaa OpenAIin, jossa se olisi yksi non-profit. Tänään hän kutsui sen hyökkäyksen.

Yesterday his lawyer kind of likened it to having a museum with a gift shop, where you wouldn't have a museum gift shop sell Picassos and pocket the profit. That's the analogy that he used. And then today we saw Elon in cross-examination, where OpenAI's lawyer was trying to get him to answer very narrow yes or no questions. And Musk was basically refusing to do so, finding ways out of it, asking for clarification and really pushing back. I mean, at one point he said,

Your questions are not simple. They're designed to trick me, essentially. And I need to give longer answers, because a simple answer would be misleading the jury. And he even used the example of, if you ask the question, have you stopped beating your wife? I can't just answer yes or no. And so they were really having this back and forth. So it got quite tetchy then. Did he persuade the jury in your assessment? What was his performance like?

I can't really speak to what the jury thinks, but he definitely was able to get his point across in his cross-examination without really giving too much away or giving any howlers that made me think, oh, he maybe shouldn't have said that. So he really made his point very clearly and then didn't really answer the questions where he could have been on the hook for something, at least in my view.

Exactly. We still have to hear from lots more people in this trial. Musk was the first person to take the stand. Naturally, we're hearing lots of his arguments for why he's bringing this case. We need to hear everybody else's arguments in defense as well. Musk is arguing that OpenAI was founded as a non-profit, which is true, with this mission of ensuring that AI benefits all of humanity. He's arguing that when he donated this money, he believed it would remain a non-profit.

By creating this hugely valuable for-profit entity now, which is worth over $850 billion, it's one of the most valuable companies in the world, Musk argues that that means that it's not going to be as focused on its mission of ensuring that AI benefits all of humanity, because instead it's going to be focused on profits. And so what is potentially at stake in the trial?

So if Musk wins, he's pushing for OpenAI to unwind its for-profit, which would make it very hard for OpenAI to then IPO, as it plans to do very soon. He's arguing that he should be having over 130 billion in damages, but he actually said that he wouldn't take that himself, he would donate it to OpenAI's charitable arm. So I think he's saying that to give himself a bit of coverage as well,

He's not trying to enrich himself, or at least he's saying he will donate the money. And he's also called for chief executive of OpenAI, Sam Altman, and OpenAI's president, Greg Brockman, to be removed from their executive positions. So that's if Musk wins.

If he loses, Musk was very inflammatory this morning when he said that it would basically cause the American charity system to fall apart, that it would set a precedent in case law for anybody to loot a charity. Technology reporter Christina Criddle.

So Anthropic didn't actually set out to build a cyber security specific model.

It was just building its next big general purpose AI model in its Claude series.

And it then discovered in testing that it was very capable at cyber security.

And in particular, it was good at detecting bugs or flaws in software.

but also this new capability of being able to generate exploits, ways to work around those bugs or flaws and exploit them.

And that's why it decided to only release it to a select amount of partners.

So AI models have been able to do this for some time now, but I think mythos really was a lightning rod.

And the idea that AI can scan for these vulnerabilities and bugs should be a positive one, right?

So you can spot where all the flaws are and then go and fix them.

But the issue is a volume problem where the AI means that

You're potentially detecting more bugs than you can solve for at a particular time.