Google DeepMind Unveils Its Most Powerful AI

Today:

GEMINI 1.0 Beats GPT-4 | AlphaCode 2 is better than 90% of coders | Gemini Nano on Pixel Phones!

Demi Saabi and the crew at Google just dropped a bomb with Gemini 1.0, and it's shaping up to be a real game-changer, going toe-to-toe with GPT-4. They're saying it's the next big thing in AI, with its ability to handle all sorts of info - text, video, you name it.

Gemini is coming in three flavors: Ultra, Pro, and Nano. Ultra's the big kahuna for complex tasks, Pro's your go-to for a variety of jobs, and Nano's perfect for on-the-go tasks, even making its way into the new Pixel phones.

WATCH THE VIDEO ON YOUTUBE

Meta AI adds Reels support and ‘reimagine,’ a way to generate new AI images in group chats, and more

Meta AI, the tech giant's AI platform, just got cooler with some new tricks. First up, there's "reimagine," a feature for group chats where you can create funky AI images using prompts. It's like a game – you start with an AI image, add your own twist with a text prompt, and voila, a new image pops up. Folks can keep adding their own prompts, making it a fun back-and-forth.

Meta AI is also getting smarter on mobile, giving better answers and summing up search results like a boss. For the Reels fans, Meta AI is now a helper. Say you're planning a trip, ask Meta AI for the coolest spots, and it'll show you Reels of those places.

These updates are hitting the streets starting today. But hold your horses, there's more in the pipeline. Meta's got over 20 new AI experiments cooking in Facebook, Instagram, and WhatsApp, covering stuff like search, ads, and chatting with businesses.

READ THE ARTICLE ON TECHCRUNCH

Apple joins AI fray with release of model framework

Apple, usually not the first one to jump into new tech trends, has rolled out some cool tools for AI. They've got this new thing called MLX, and another one named MLX Data. These are like building blocks for folks who want to create AI stuff that works great on Apple devices.

So, MLX is this framework where tech wizards can whip up models that run super smooth on Apple's own chips. And MLX Data? That's a library for deep learning models. Both of these goodies are up for grabs on GitHub and PyPI.

MLX is inspired by some big names in the AI world but has a unique twist. It's got shared memory, which means it can handle tasks on different devices without having to shuffle data around. Word is, it's simple enough for newbies but still packs a punch for training complex AI models.

READ THE ARTICLE ON THE VERGE

BlackRock to roll out first generative AI tools to clients next month

BlackRock, this huge money handling giant with a whopping $9.1 trillion under management, is getting ready to launch its first AI tools next month. They're calling it a "co-pilot" for their risk management systems, Aladdin and eFront. Basically, their clients can use these fancy AI language models to sift through heaps of info more easily.

The big deal here is that BlackRock's already raking in over 8% of its cash from tech services, and with this move, they're looking to stay ahead in the game. They've got big plans to use AI for stuff like helping their investment pros gather data for reports and investment pitches, and even a language translator. Plus, they're planning to roll out Microsoft's AI add-on for Office 365 across the company.

BlackRock's bigwigs believe this AI strategy will help them cut costs and boost their profit margins. They're thinking that these large language models and automation can make their team more productive, pushing those gains into 2024 and beyond.

READ THE ARTICLE ON FINANCIAL TIMES

Liquid AI, a new MIT spinoff, wants to build an entirely new type of AI

Liquid AI, a new kid on the block from MIT, is cooking up a fresh type of AI. They're backed by a heavy-hitting $37.5 million and valued at a cool $303 million. The brains behind this are Daniela Rus and some sharp folks from MIT who've been deep in the AI game.

They're all about "liquid neural networks." Imagine an AI brain that's super flexible, kinda like how a worm thinks. These networks are tiny compared to regular AI models and don't need a ton of computer power to run. We're talking about something so light, it could run on a gadget as simple as a Raspberry Pi.

The team at Liquid AI trained their system with data from a pro drone pilot and then let it loose on some drones. The results? Their liquid brain could navigate drones through tricky spots like forests and city streets, even better than other AI models.

Now, the big goal for Liquid AI is to turn this tech into something businesses can use. They're aiming to outdo the big names in AI, offering a platform for companies to build their own models for whatever they need. The plan is to keep making these liquid neural networks better and more practical for real-world use.

READ THE ARTICLE ON TECHCRUNCH

McDonald’s will use Google AI to make sure your fries are fresh, or something?

McDonald's is teaming up with Google to bring AI into their joints starting 2024. They're planning a big makeover with new tech in loads of their stores. The big idea? They say it's all about serving up hotter, fresher grub like fries straight out of the fryer. 

The plan includes both new gear and software in the stores, plus some fancy stuff from Google Cloud. They're hoping this tech will help the store bosses spot and fix problems faster, making everything run smoother.

Now, McDonald's is keeping it hush-hush about whether this AI move means fewer jobs for folks. They're talking a big game about making things easier for their staff and spicing up the customer experience. Wendy's, another fast-food giant dipping its toes in AI, is playing the same tune. They've been testing AI for taking orders, but it's still early days.

READ THE ARTICLE ON THE VERGE

Meta and Microsoft say they will buy AMD's new AI chip as an alternative to Nvidia's

Meta, OpenAI, and Microsoft are shaking things up by choosing AMD's new AI chip, the Instinct MI300X, over Nvidia's usual go-to chips for AI stuff. This big move happened at an AMD event, where they showed off this new chip. It's a big deal 'cause Nvidia's chips are kind of the top dog in the AI world, but they're pricey.

AMD's new chip is supposed to be a game-changer. It's built on some fresh architecture, which usually means better performance, and it's packed with 192GB of some high-speed memory called HBM3. This lets it handle bigger AI models and work faster.

Lisa Su, the big boss at AMD, says their chip will make AI responses quicker and smoother. But there's a catch. Switching from Nvidia to AMD isn't a walk in the park. Companies gotta put in time and money to make the switch. AMD's trying to sweeten the deal by improving their software and making it more competitive with Nvidia’s.

Who's lining up for this chip? Meta and Microsoft, two heavy hitters who used to buy a lot of Nvidia's stuff, are now going for AMD’s MI300X. Meta plans to use it for AI tasks like smart stickers and editing images. Microsoft will offer it through their Azure web service, and Oracle's cloud service is jumping on board too. OpenAI's also getting in on the action by supporting AMD GPUs in one of their software products, Triton.

READ THE ARTICLE ON CNBC

AI accelerates problem-solving in complex scenarios

Researchers from MIT and ETH Zurich have used machine learning to make solving complex problems faster and more efficient. These problems, like optimizing delivery routes for companies like FedEx during the holiday season, are so complicated that they require specialized software. This software, called a mixed-integer linear programming (MILP) solver, can take a long time to find the best solution.

The researchers found a way to speed up MILP solvers by simplifying a critical step and using machine learning to find the optimal solution for specific problems. This approach can be applied in various fields, including ride-hailing services, electric grids, and vaccination distribution.

Their method improved the speed of MILP solvers by 30 to 70 percent without sacrificing accuracy. It allows companies to customize the solver for their specific needs using their own data.

READ THE ARTICLE ON MIT