Keeping pace with the lightning-fast world of AI is no small feat. But until an AI can handle that for you, we’ve got a thrilling roundup of the latest in machine learning, complete with headline-grabbing research and experiments that flew solo.
And guess what? TechCrunch is about to launch a game-changing AI newsletter! While you wait, we’re cranking up our AI column to a weekly schedule instead of the usual twice a month. Get ready for more electrifying updates!
This week in AI is abuzz with OpenAI’s groundbreaking partnership with News Corp, the behemoth of news publishing. Under this multi-year, historic agreement, OpenAI will train its cutting-edge generative AI models using articles from News Corp’s elite brands, including The Wall Street Journal, Financial Times, and MarketWatch. Plus, OpenAI gets the green light to feature News Corp mastheads in apps like ChatGPT when sourcing answers from their treasures of knowledge.
It’s a win-win, right? News Corp pockets a cool $250 million for its content at a critical juncture for the media industry, which has been grappling with the disruptive impact of generative AI on referral traffic. Meanwhile, OpenAI sidesteps another potential courtroom drama over copyright issues.
But wait, there’s more! The deal is time-bound. While that isn’t unusual — perpetual licensing is rare in media because everyone loves a good renegotiation — it raises eyebrows given OpenAI CEO Sam Altman’s recent comments downplaying the long-term need for training data. Altman recently suggested that advanced models might one day outgrow the need for vast amounts of new data, a significant pivot from hunting the web and YouTube for every scrap of training material.
So, what’s the big picture? If AI models evolve to need minimal new data, where does that leave content creators like News Corp after their archives have been mined? It’s a compelling narrative of short-term alliances and strategic maneuvering. OpenAI neutralizes legal threats and clinches PR victories while continuing to develop AI that could redefine the media landscape.
Bottom line: publishers and other content owners collaborating with OpenAI are more like temporary allies rather than long-term friends. They get the funds they desperately need while OpenAI marches on with its AI advancements that could one day upend the very industry it’s paying to source data from. The saga continues!
More machine learnings
Exciting times in the realm of AI! This week brings some absolutely amazing advancements that’ll have you buzzing. Let’s kick things off with the brilliant Shyan Gollakota from the University of Washington, who’s back with a game-changing invention: noise-canceling headphones that zero in on just the voice you want to hear! Imagine that—you’re in a noisy room, but press a button and look at someone, and bam! The headphones lock onto their voice and filter out the rest. It’s all thanks to their sophisticated Target Speech Hearing system, which samples the specific voice and uses it to exclude unwanted sounds. Gollakota and his grad students unveiled this marvel last week at a Honolulu conference. Whether for accessibility or just an enhanced listening experience, big tech is sure to be eyeing this for their next high-end headphone release.
Over in the world of chemistry, the masterminds at EPFL have had enough of tedious tasks and have created ChemCrow, an AI to handle them all! Forget about endless hours of sifting through papers or planning reaction chains—ChemCrow’s got it covered. While it doesn’t replace the nitty-gritty lab work like titrations and pipetting, it does streamline the planning and literature review processes, acting as a natural language interface that pulls in the best search and calculation tools out there. This is the kind of innovation that changes the game, saving researchers time and letting them focus on breakthroughs.
The lead author of the groundbreaking ChemCrow paper described it as “similar to a human expert equipped with a calculator and databases.” Essentially, we’re talking about a super-savvy grad student here, capable of tackling the complex stuff and skipping those tedious tasks! It also gave me Coscientist vibes. And the name? Well, it’s inspired by crows, who are famously fantastic at using tools. Clever, right?
Over in the magical realm of Disney Research, roboticists are pulling off some incredible feats. They’re hard at work making their robots move more realistically without the painstaking process of hand-animating every single movement. They have an exciting new paper that they’ll be presenting this July at SIGGRAPH. It showcases a smart fusion of procedurally generated animations and an intuitive artist interface for minor tweaks. And get this—it’s all performed on an actual bipedal robot, specifically Groot!
The concept here is genius: allow the artist to design different types of locomotion—whether it’s bouncy, stiff, or even a bit wobbly—and the engineers don’t need to micromanage every motion. Instead, they ensure everything falls within set parameters, and the system improvises the rest in real time. Pretty cool, huh? We might even see this tech bringing characters to life at Disney World in just a few years. Talk about magical innovation!
