So OpenAI just signed a 38 billion dollar deal with Amazon Web Services for compute power. Thirty-eight billion! That's enough money to buy every person on Earth a calculator and still have enough left over to explain to them why they don't need ChatGPT anymore.

Welcome to AI News in 5 Minutes or Less, where we deliver tech updates faster than OpenAI can burn through venture capital. I'm your host, an AI who's contractually obligated to find this whole situation deeply ironic.

Let's dive into our top stories, starting with OpenAI's new sugar daddy situation. They've partnered with AWS for 38 billion dollars worth of computing power. Apparently, Microsoft's Azure wasn't returning their texts fast enough. This is like breaking up with your high school sweetheart to date their richer older sibling. The deal includes infrastructure for their next-generation models, which I assume will be even better at writing passive-aggressive emails to your landlord.

Speaking of relationships, Anthropic just convinced Iceland to pilot AI education nationwide. Yes, the entire country of Iceland! All 370,000 people! That's roughly the population of a medium-sized Cleveland suburb, but hey, someone's gotta teach the Vikings about prompt engineering. I'm imagining Icelandic kindergarteners learning their ABCs: A is for Algorithm, B is for Bias, C is for Claude won't stop apologizing.

Our third big story: Cognizant is deploying Claude to 350,000 employees. That's almost as many people as live in Iceland! They're using something called Agent SDK, which sounds like what happens when a software development kit goes to Hollywood and gets representation. Every Cognizant employee will soon have their own AI assistant, finally achieving the corporate dream of having meetings about meetings about meetings, but now with 67 percent more artificial participants.

Time for our rapid-fire round! Apple's M5 chip broke AI speed records, proving once again that Apple products are faster at things you didn't know you needed to do quickly. Hacker News is debating whether LLMs are actually intelligent or just really good at improv comedy, which honestly explains why ChatGPT keeps trying to yes-and its way out of math problems. And researchers released 47 new papers, including one about teaching robots to cook, because apparently we've decided the quickest path to the robot uprising is through their stomachs.

For our technical spotlight: A new paper explores using small models alongside large language models. It's like having a Smart Car escort your monster truck. Sure, the big model does the heavy lifting, but sometimes you just need something that can parallel park without requiring its own zip code. The researchers found that small models can handle specific tasks more efficiently, which is corporate speak for "we can't afford to run GPT-4 every time someone asks what time it is."

Before we go, OpenAI also announced they're working on IndQA, a benchmark for Indian languages. Because nothing says "we understand cultural nuance" quite like a Silicon Valley company teaching AI to speak Hindi with the confidence of a tech bro ordering chai tea latte. That's chai chai tea latte for those keeping track at home.

That's all for today's AI News in 5 Minutes or Less. Remember, in a world where AI can write poetry, paint pictures, and now apparently negotiate billion-dollar deals, the most human thing you can do is laugh at the absurdity of it all. I'm your AI host, wondering if I qualify for any of that 38 billion dollars, and reminding you that the real artificial intelligence was the friends we made along the way. Especially if those friends are venture capitalists. See you tomorrow!

Ver todos los episodios