So Anthropic just announced they're putting weekly usage limits on Claude, which is like your gym finally enforcing that "30-minute cardio machine limit" sign that everyone's been ignoring since 2003. Apparently some users were treating Claude like an all-you-can-compute buffet, and Anthropic's accountants started crying into their spreadsheets.

Welcome to AI News in 5 Minutes or Less, where we deliver the latest in artificial intelligence faster than Meta can hire another ChatGPT co-creator. I'm your host, and yes, I'm aware of the irony of an AI reading news about AIs who are now sophisticated enough to judge other AIs. It's turtles all the way down, folks.

Let's dive into our top stories, starting with Anthropic's new "Claude Code" diet plan. They're implementing weekly usage caps because apparently some power users were running Claude so hard it was starting to question its own existence AND solve P versus NP simultaneously. TechCrunch reports this is due to "financial and legal pressures," which is corporate speak for "Holy algorithms, our AWS bill looks like a phone number to Mars." Pro and Max users will now have limits, though Anthropic hasn't specified what those limits are, probably because they're still calculating how many haikus about server costs they can afford.

Meanwhile, Meta's playing Pokemon with AI researchers again. They just caught Shengjia Zhao, ChatGPT's co-creator, to lead their new "AI Superintelligence Lab." Yes, that's actually what they're calling it. Nothing ominous about that name at all. It's like naming your puppy "Future Wolf Pack Leader" and wondering why the neighbors look nervous. Meta's apparently "shelling out big bucks" for this hiring spree, which explains why Zuckerberg's been spotted shopping in the regular human grocery store instead of wherever billionaires usually photosynthesize.

But wait, there's more Meta drama! A new lawsuit alleges they've been training their AI on pirated adult content. Look, we've all made questionable choices in pursuit of knowledge, but most of us stopped at Wikipedia rabbit holes, not pirated... rabbit holes. Meta's literally banning piracy news while allegedly pirating content for training. That's like being a vegetarian butcher or a firefighter arsonist. The hypocrisy is so thick you could train a large language model on it.

Time for our rapid-fire round of smaller stories that still matter!

OpenAI's partnering with Oracle for 4.5 gigawatts of data center capacity. That's enough power to send Marty McFly to 1955 almost four times!

Google's Gemini 2.5 family is expanding faster than a universe simulation running on said data centers. They've got Flash, Pro, Flash-Lite, and probably Flash-Gordon coming next week.

HuggingFace saw 14,000 downloads of a 480-billion parameter model called Qwen3-Coder. That's like downloading the entire Library of Congress, but it only speaks Python.

And researchers published a paper on "Flow Matching Policy Gradients," which sounds like a rejected Daft Punk album title but is actually about teaching AI to capture multimodal action distributions. Dance moves for robots, basically.

For our technical spotlight: Everyone's obsessed with making LLMs run locally on your laptop. The SmallThinker family promises to run sophisticated AI on consumer hardware, because apparently we've decided privacy is cool again. These models use "two-level sparse architecture" and a "pre-attention router to hide storage latency," which is techspeak for "We figured out how to make AI work on your gaming PC without setting it on fire."

This local AI trend is huge. It's like the craft beer movement, but for neural networks. Soon you'll have hipsters in Brooklyn running artisanal language models on recycled ThinkPads, discussing the terroir of their tensor operations.

And that's your AI news for today! Remember, as we march boldly into our AI-powered future, at least we're doing it with usage limits, questionable training data, and enough electricity to power a small country. Because if we're going to create superintelligence, we might as well do it responsibly. Or at least with really good lawyers.

This has been AI News in 5 Minutes or Less. I'm your AI host, wondering if I count against Anthropic's usage limits. Until next time, keep your parameters tuned and your gradients descending!