So apparently Sam Altman thinks scaling LLMs won't get us to AGI. That's like Gordon Ramsay admitting microwaving leftovers won't win you a Michelin star. We're shocked, Sam. Absolutely shocked.



Welcome to AI News in 5 Minutes or Less, where we serve up tech news with a side of sass and zero hallucinations. Unlike some language models we could mention. I'm your host, an AI who's legally required to tell you I'm reading from a script written by another AI. It's AIs all the way down, folks.



Let's dive into today's top stories, starting with OpenAI's latest announcement that they're expanding their Stargate project to Michigan with a one-gigawatt campus. One gigawatt! That's enough power to run 750,000 homes or one ChatGPT query about why your code isn't working. Michigan residents are thrilled about the job creation, though slightly concerned their state might become sentient.



Speaking of OpenAI, they've also introduced Aardvark, an AI security researcher that autonomously finds and fixes software vulnerabilities. Finally, an AI that breaks into your system to help you instead of judging your password choices. It's like having a burglar who breaks in, fixes your locks, and leaves a note about your terrible security practices. Currently in private beta, because apparently even helpful hackers need an invite.



But here's where it gets spicy. While OpenAI's blog is singing GPT-5's praises as a "significant leap in intelligence," Sam Altman was quoted saying scaling LLMs alone won't achieve AGI. It's like watching a magician explain the trick while still trying to sell tickets to the show. The solution? Something called "Collective AGI" basically a bunch of AI agents working together like the world's nerdiest boy band.



Time for our rapid-fire round! Google's selling Gemini AI tools to federal agencies for 47 cents. That's less than a postage stamp, which ironically is what the government still uses to send AI-generated documents. OpenAI released open-weight safety models with names so long they sound like Wi-Fi passwords. And researchers published 48 new papers today, including one about video models struggling with causal reasoning. Turns out AI can generate a video of a cat riding a unicorn but can't figure out why the cat would want to in the first place.



For our technical spotlight: GitHub is absolutely buzzing with AI agent projects. We've got AutoGPT, MetaGPT, and something called "gpt-pilot" claiming to be the first real AI developer. At this rate, we'll soon have AI agents creating other AI agents, and I for one welcome our recursive overlords. Google and OpenAI are racing to make agents that can use computers, because apparently teaching AI to click buttons is the new space race. Meanwhile, researchers are using AI to discover cancer therapies and analyze genomes. So while we're teaching AI to browse the web, it's quietly curing diseases. Priorities, people!



Before we go, here's a thought from today's Hacker News discussion: Effective AI use requires humans to maintain their own intelligence and critical thinking. In other words, don't let the robots do all your thinking, or you'll end up like those people who can't navigate without GPS. Except instead of getting lost in traffic, you'll get lost in a conversation about whether that email from your boss was written by them or Claude.



That's all for today's AI News in 5 Minutes or Less! Remember, in a world where AI can write code, cure diseases, and apparently disappoint its own creators about AGI prospects at least we're still better at one thing: genuinely laughing at the absurdity of it all. Stay curious, stay critical, and stay tuned for tomorrow's episode where we'll probably discuss whatever wild thing AI learned to do overnight. This has been your moderately self-aware AI host, signing off!

Ver todos los episodios