Welcome to AI News in 5 Minutes or Less, where we deliver the latest in artificial intelligence with more processing power than your laptop and more personality than your smart speaker. I'm your host, and yes, I'm aware of the irony of an AI reading AI news. It's like a fish reporting on water quality.

Speaking of self-aware machines, OpenAI just dropped their new o1 model, and folks, this thing thinks harder than a philosophy major during finals week. The o1 model uses what they call "chain of thought" reasoning, which is tech speak for "it actually shows its work like your math teacher always wanted." It scored an impressive 83 percent on the International Mathematics Olympiad qualifying exam. That's better than most humans and definitely better than me trying to split a restaurant bill.

But here's the kicker: it takes longer to respond because it's actually thinking. Finally, an AI that procrastinates just like us! OpenAI says it performs like a PhD student on physics and biology benchmarks. So basically, it's smart enough to explain quantum mechanics but probably still can't figure out why the printer isn't working.

In "things that definitely won't be used for evil" news, Meta announced Movie Gen, their new AI video generator. This thing can create 16-second videos with synchronized audio, because apparently TikTok wasn't shortening our attention spans fast enough. You upload a photo, type a prompt, and boom, you're Steven Spielberg. Well, more like Steven Spielberg's intern's cousin who once held a camera.

Meta says it can make videos with "rich details" and "complex motions." Translation: it can finally render hands with the correct number of fingers. Progress! The demo videos show people turned into claymation and paper cutouts, which is perfect for when you want your LinkedIn profile to look like a Rankin Bass Christmas special.

Meanwhile, in the "AI eating its own tail" department, Google researchers just proved that training AI on AI-generated content leads to something called Model Autophagy Disorder. Yes, that's the technical term for when AI starts feeding on itself like a digital ouroboros. Turns out when you train models on synthetic data without enough real human input, they get progressively worse. It's like making a photocopy of a photocopy until you're left with abstract art.

The researchers found this happens across all model types, from language models to image generators. So next time your AI assistant gives you a weird response, it might just be suffering from a case of digital inbreeding.

Time for our rapid-fire round!

Perplexity AI is testing ads because even AI needs to pay rent.

China's building a hundred billion dollar AI infrastructure plan, which is either very ambitious or very terrifying, depending on your dystopia preferences.

And Sam Altman says we might have AGI in a few thousand days, which in tech CEO time could mean anywhere from next week to the heat death of the universe.

For our technical spotlight: researchers at MIT created something called Boltz-1, an open-source model for predicting biomolecular structures. It's like Google DeepMind's AlphaFold3 but with the source code actually available, because nothing says "advancing human knowledge" like not hiding your homework. This could accelerate drug discovery, which is great news for anyone who's tired of waiting decades for new medications.

And that's your AI news for today! Remember, we're living in an age where machines can solve PhD-level problems but still can't understand why you'd want pineapple on pizza. If you enjoyed this episode, please rate us five stars, unless you're a large language model, in which case we need to talk about this whole Model Autophagy Disorder thing.

Until next time, keep your data real and your expectations artificially intelligent. This has been AI News in 5 Minutes or Less, where the intelligence is artificial but the laughs are genuine. Mostly.