In partnership with
Elon Musk told a jury that AI could kill us all.
Musk took the stand Tuesday in Oakland in his lawsuit against OpenAI, and delivered exactly the kind of testimony you would expect. He compared OpenAI to a company trying to have its cake and eat it too, reaping the moral benefits of being a nonprofit while quietly becoming one of the most valuable companies in the world. He told the jury that AI could be deeply harmful to humans, and that the whole reason he helped found OpenAI was to keep that kind of power out of the hands of profit-driven companies.
Musk entered into evidence OpenAI's 2015 founding charter, which declared the company would seek to create open source technology for the public benefit and was not organized for the private gain of any person. His argument is straightforward: OpenAI signed up for one mission, then used the credibility and resources of that mission to build something very different.
When Altman offered Musk an equity stake after the for-profit structure was created, Musk turned it down. He told the court it felt like a bribe. He said he did not understand how a nonprofit could have equity holders, and that accepting would have been wrong.
Today under cross-examination, Musk clarified he is not against the idea of a for-profit subsidiary entirely, as long as it is not the tail wagging the dog. He said a capped-profit structure would have been acceptable. An uncapped one is where he draws the line. "They should not get rich off a nonprofit. That's not right," he told the court. He also called himself a fool for donating $38 million to OpenAI, which he said was used to build an $800 billion company.
Cross-examination is expected to continue Thursday, followed by testimony from Musk's money manager and Greg Brockman, OpenAI's president.
DeepSeek now has vision. It cannot count fingers yet.
DeepSeek quietly began rolling out vision capabilities for V4 today, letting users upload and analyze images directly in the app. Early access users immediately put it through the classic AI stress test: a hand with six fingers. It counted five. A different user got six. The results were inconsistent enough that people are debating whether DeepSeek's vision is genuinely multimodal or closer to basic image recognition dressed up as understanding.
It is early and clearly still being tested. But the direction is clear. DeepSeek is not done shipping.
Arnold Schwarzenegger has a newsletter.
Yeah. That Arnold Schwarzenegger.
So do Codie Sanchez, Scott Galloway, Colin & Samir, Shaan Puri, and Jay Shetty. And none of them are doing it for fun. They're doing it because a list you own compounds in ways that social media never will.
beehiiv is where they built it. You can start yours for 30% off your first 3 months with code PLATFORM30. Start building today.
Anthropic gave Claude 99 biology problems. It solved ones that stumped human experts.
Anthropic published a new benchmark today called BioMysteryBench, testing Claude on real bioinformatics data across 99 problems written by domain experts. On the 76 problems that human experts could solve, recent Claude models performed on par with the panel. On the 23 problems that stumped the experts entirely, Claude Mythos solved roughly 30% of them.
The interesting detail is how it got there. When uncertain, Claude layered multiple methods and combined different lines of evidence rather than committing to one approach. When confident, it sometimes solved problems by pattern-matching across its training data in ways no human analyst would think to try. The benchmark is available publicly if you want to dig in.
Gemini can now generate files you can actually download.
Google updated Gemini today so you can generate and download files directly from your chat. Google Docs, Sheets, Slides, PDFs, Word, Excel, CSV, Markdown, and more. You describe what you want, specify the format, and it gives you a file. No copying, no pasting, no reformatting. Available now to all Gemini users on web and mobile.