DeepSeek Is Back With a New Flagship
A year after DeepSeek rattled Silicon Valley with a model that punched way above its weight on a fraction of the compute budget, they just dropped another one. They're calling it their most powerful open-source model yet, and it's being pitched as a direct challenge to OpenAI and Anthropic.
Here's what matters: DeepSeek keeps proving that you don't need a $10 billion training run to build a world-class model. Every time they ship something, it puts pressure on American labs to justify their spend. Open-source getting this competitive this fast is good for developers and bad for anyone whose moat is just "we have more GPUs."
I'm downloading it this weekend. If it's half as good as the hype, it'll be my new local workhorse.
A Lawyer Got Suspended for 57 Bad AI Citations
The Nebraska Supreme Court just suspended attorney Greg Lake after his appellate brief had 57 defective citations out of 63. Twenty of them were straight-up AI hallucinations. Made-up cases. Cases that don't exist.
This isn't the first time and it won't be the last. U.S. courts handed out at least $145,000 in sanctions against attorneys for AI citation errors in just the first quarter of this year. That's a lot of lawyers learning the same expensive lesson.
Look, I'm not anti-AI in legal work. But if you're going to hand a brief to a judge, you have to verify every single citation yourself. Full stop. AI is a research assistant, not a paralegal you can just trust. The tool doesn't know what it doesn't know, and it will confidently make stuff up. That's just how these models work right now.
Use it to draft, use it to brainstorm. But verify before you file.
Oracle Is Cutting 30,000 Jobs to Pay for AI Infrastructure
Oracle announced it's cutting between 20,000 and 30,000 employees so it can redirect $8 to $10 billion toward AI infrastructure. Block cut 4,000 people, nearly 40% of its workforce, with CEO Jack Dorsey saying outright that AI made those roles redundant.
This is the part of the AI story that doesn't get enough honest coverage. The productivity gains are real, but they're not being spread evenly. A new PwC study out this week found that three-quarters of AI's economic gains are going to just 20% of companies. The winners are winning big and the rest are still figuring out where to start.
I'm not doom-posting here. I think there's real opportunity for people who adapt. But the timeline is faster than most people expected, and "we'll figure it out eventually" isn't a plan.