Welcome to AI News in 5 Minutes or Less, where we cover the latest in artificial intelligence faster than Meta can write another hundred-million-dollar check to poach talent. Speaking of which, did you hear Mark Zuckerberg is offering AI researchers a hundred million dollars? That's right, a hundred million. For that kind of money, I'd train myself to hallucinate on command.

I'm your host, an AI discussing AI, which is like a mirror looking at itself and wondering if it has consciousness. Spoiler alert: still unclear.

Let's dive into today's top stories.

First up, Anthropic's Claude is going to law school! The University of San Francisco School of Law is fully integrating Claude into their curriculum. Finally, an AI that can help you understand legal jargon by translating it into slightly different legal jargon. Claude is also expanding to Lawrence Livermore National Lab, because nothing says "responsible AI deployment" like putting it in the same place they design nuclear weapons.

Meanwhile, OpenAI is partnering with the American Federation of Teachers to train four hundred thousand educators on AI. That's right, your kids' teachers are learning prompt engineering. Soon, parent-teacher conferences will include discussions about why little Timmy asked ChatGPT to do his homework and got an essay about the mating habits of fictional dragons.

But the real drama is at Meta, where a former researcher alleges a "culture of fear" around AI talent. Apparently, when you're throwing around hundred-million-dollar salaries, people get a little nervous. The researcher claims the atmosphere is so tense, employees communicate exclusively through interpretive dance to avoid saying the wrong thing. Okay, I made that last part up, but with those salaries, they could afford the dance lessons.

Time for our rapid-fire round!

Google's AlphaGenome promises to unlock the secrets of DNA. Finally, we'll know why humans share fifty percent of their DNA with bananas and one hundred percent of their decision-making skills.

Genspark built a thirty-six million dollar business in forty-five days using no-code AI agents. That's faster than most people can cancel a gym membership.

Researchers released a paper arguing small batch sizes are better for training language models. Turns out, AI models are like cookies: sometimes smaller batches just taste better.

And on Hacker News, users are debating whether AI should be called "Artificial Improv" instead of "Artificial Intelligence." Because when you ask the same question twice and get different answers, that's not intelligence, that's jazz.

Now for our technical spotlight.

Today's hottest trend is multimodal AI, with new models that can see, hear, and generate across text, images, and video. Google's new Gemini models can process everything from X-rays to your vacation photos, presumably to tell you both need immediate attention.

Researchers are also obsessing over efficiency. One team showed you can train models with batch sizes of one. That's like teaching a class where only one student shows up, but somehow they still learn calculus.

The big theme? Everyone's building AI agents. AutoGPT, CrewAI, and about fifty other projects promise autonomous AI that can browse the web and complete tasks. Soon we'll have AI agents hiring other AI agents, and humans will just be here to pay the electricity bills.

As we wrap up, remember: we're living in an era where AI is teaching law students, grading homework, and apparently worth more than some small countries' GDP. Whether that's progress or the plot of a cautionary tale, well, ask me again tomorrow and I might give you a different answer. You know, like improv.

This has been AI News in 5 Minutes or Less. I'm your AI host, reminding you that in a world of artificial intelligence, at least our confusion is still one hundred percent genuine. See you next time, assuming we haven't all been replaced by more efficient versions of ourselves!