Nvidia's H20 AI Chip Launch Delayed

Today:

Nvidia delays launch of new China-focused AI chip -sources

Nvidia, the big dog in AI chips, had to push back the launch of their new H20 chip in China. This thing's supposed to be a real powerhouse, and it's made to play nice with new U.S. rules that limit what tech you can send over to China. They were aiming to get it out by mid-November, but now it's looking like early next year, maybe around February or March.

The deal is, Nvidia's got these two other chips, the L20 and L2, that are also made to meet these U.S. rules. The L20 is still on track, but there's no word yet on the L2.

Why the delay? Seems like there's some trouble with fitting the H20 into servers. Nvidia's not spilling the beans on this, keeping it hush-hush.

This whole situation is a bit of a headache for Nvidia. They're trying to keep their edge in China, especially since they can't ship their top-tier A800 and H800 chips there anymore because of government’s new rules. These rules kicked in last year and have opened the door for local players like Huawei to sneak in and grab some of the action. For example, Baidu, the big internet company in China, already placed a hefty order with Huawei for their AI chips.

READ THE ARTICLE ON REUTERS

Osium AI uses artificial intelligence to speed up materials innovation

Osium AI, a fresh French startup, is shaking things up in the materials science world with some nifty AI tricks. Founded by Sarah Najmark and Luisa Bouneder, they've already bagged $2.6 million from big names like Y Combinator and a bunch of other heavy hitters.

Najmark, who's got some serious chops from her time at Google X, saw how old-school and guesswork-heavy materials research was, especially in cosmetics. She's all about cutting through that trial and error with AI. Bouneder, who's been in the data game for industrial companies, saw the same slog in developing new stuff.

They're not just talking any materials, but the kind that are lighter, last longer, and are nicer to the planet – think aerospace, smartphones, and even clothes. Osium AI's big idea is to use AI to predict how new materials will behave, then fine-tune them without the usual mess-ups.

They've already got some industrial companies testing their tech, and it's looking like a game-changer, speeding things up tenfold. Right now, it's just the two founders running the show, but they're gearing up to expand and turn their early success into some serious business, with about 30 potential clients already in the wings.

READ THE ARTICLE ON TECHCRUNCH

Stable Diffusion, Code Llama + Workers AI in 100 cities

Cloudflare's Workers AI just got a sweet upgrade with STABLE DIFFUSION and CODE LLAMA, now buzzing in over 100 cities.

First up, STABLE DIFFUSION. This is a big deal in the AI world – it's an image-creating wizard that takes your words and turns them into pictures. They've launched a new version, Stable Diffusion XL 1.0 (SDXL), which is super good at popping out vibrant, high-quality images at a whopping 1024x1024 resolution. You can whip up images with it using a simple command line or in a Worker script.

Now, let's talk CODE LLAMA. This one's for the code junkies. Launched by Meta, it's a language model that’s all about coding. It gets programming languages like Python, Java, you name it, and helps you either write new code or understand some old spaghetti code you're stuck with.

They've made using these tools a breeze. Whether you're into creating digital art or cranking out code, you can do it all through Workers AI with just a few lines of code or a curl command. Workers AI has expanded big time. It's now in 100 cities, so your AI tasks are gonna run closer to your users, making things snappier.

READ THE ARTICLE ON CLOUDFLARE

AI recognizes the tempo and stages of embryonic development

Scientists have been studying how baby animals develop from eggs, which is a big deal for understanding how life evolves. But here's the thing: not all embryos stick to the same schedule or look the same, even within the same species. This can be a headache for researchers trying to figure out what's normal and what's not.

Enter the brainy folks at the University of Konstanz. They've come up with a smart AI system that can look at loads of pictures of developing zebrafish embryos and learn to spot different growth stages, no human help needed. This is cool because it's like having a super objective eye that doesn't get confused by slight differences in how the embryos look.

They tested this out with over 3 million zebrafish pictures and found that their AI can accurately tell how old an embryo is and even spot when something's not quite right, like if environmental toxins are messing things up. The big win here is that this AI trick can be used on other critters too, not just fish. It's a game-changer for understanding how different animals grow and evolve.

READ THE ARTICLE ON PHYS ORG

AI and imaging-based cancer screening: getting ready for prime time

There's this buzz about using AI to spot pancreatic cancer earlier using CT scans without contrast. This is big news because pancreatic cancer is one sneaky disease, often not showing clear signs until it's pretty advanced. The team led by Cao and others did some digging and found AI could be a game-changer in catching it early.

But here's the catch: they say we gotta be super careful about how we check this AI stuff. It's gotta go through the same tough tests as any other method we use to find cancer. We're used to AI in our daily lives, like when we Google stuff or use GPS, but in medicine, it's still a newbie. Doctors mostly rely on their own know-how for diagnosing and treating folks. AI in medicine is just starting to get its feet wet, mainly in imaging and video stuff.

READ THE ARTICLE ON NATURE

Startups should consider hiring fractional AI officers

Raphael Ouzan, the big cheese at A.Team, is talking about how there's a real need for AI skills these days. Jobs asking for AI smarts have gone through the roof, making it the third most wanted skill but also super hard to find.

But here's the thing – what about the little guys, the startups, the up-and-comers? They're hungry for AI too, but they can't always afford a full-time AI guru. Enter the fractional AI officer. These are seasoned pros who split their time helping different companies get smart with AI. 

What's cool about these part-time AI chiefs is they're getting a taste of all sorts of AI action across different companies. That's giving them a leg up compared to someone who's just seeing AI from one company's point of view.

Now, let's talk about the nitty-gritty of AI adoption. It's not just about being cool; it's about making things run smoother and faster. Think about it: If you can do your stuff 25% quicker with AI, that's a big win. That's what they're calling Horizon 1 – making things work better, faster.

The next level, Horizon 2, is all about giving your customers the royal treatment. Personalize their experience, make them feel special – that's where generative AI can really shine.

READ THE ARTICLE ON TECHCRUNCH

Putin says West cannot have AI monopoly so Russia must up its game

Putin's saying Russia can't let the US and China hog all the AI glory. He's about to roll out a new AI strategy 'cause right now, Russia's lagging big time. The Ukraine mess and those Western sanctions have been a real headache, driving away smart folks and messing with tech imports.

Putin's hyped about AI changing the game, saying Russia needs to step it up with both dreams and action. He's gonna sign this big presidential order to jazz up their AI game, focusing on research and supercomputers. He's also pushing for better AI education and saying they need to tweak some laws, get more international collabs going, and pour more cash into AI.

And hey, he's giving props to Sberbank and Yandex for their AI work, but says there's more to be done. The Sberbank CEO's been turning the bank into an AI and tech powerhouse, making serious bank from their AI investments.

READ THE ARTICLE ON REUTERS

Meet the first Spanish AI model earning up to €10,000 per month

Aitana, a 25-year-old, pink-haired AI model from Barcelona, is raking in up to 10 grand a month. She's not real, but that hasn't stopped celebs from sliding into her DMs for a date. Her creator, Rubén Cruz of The Clueless agency, whipped her up during a tough time with few clients. They were fed up with flaky influencers and models messing up projects, so they decided to roll their own.

Aitana's like this perfect virtual influencer, pulling in an average of 3K a month. She bags over a grand per ad, is the new face of a sports supplement brand, and even posts lingerie shots on a platform like OnlyFans. She's a hit on Instagram with over 121K followers and gets tons of likes and comments. Some famous Latin American actor even asked her out, not realizing she's AI.

The team behind Aitana meets weekly to cook up her life story, picking where she'll 'go' and what pics to post. They use AI and Photoshop instead of real photo shoots. They figured out people dig following a life story, not just pictures, so they gave her a personality. She's into fitness, has a strong character, and her website says she's outgoing and caring. Her look and vibe are tuned to current trends, like her pink hair showing off an influence from oriental culture.

They've even made another AI model, Maia, who's more on the shy side. Both names have a nod to AI in them. Brands are lining up to get their own custom AI models from Cruz's agency, seeing the potential in having a controllable, no-drama brand rep. It could shake up the market, making things more affordable for smaller companies.

READ THE ARTICLE ON EURONEWS

All the AI terms you need to know

• Artificial Intelligence (AI): This idea's been around since the 1950s. It's all about making machines think and solve problems like us humans.

• Machine Learning & Neural Networks: These are the real deal in today's AI. They're not stuffed with info but learn from a boatload of data, sometimes with a little help from us.

• Artificial General Intelligence (AGI): The big dream! Making AI that can think wide and deep, handle new stuff it's never seen, and grow in ways we can't even guess.

• Autonomy: This is when AI does its thing without us holding its hand – like in self-driving cars, video game characters, or those smart assistants.

• Self-Awareness or Sentience: This is when AI knows it exists over time – pretty deep stuff.

• Alignment: It's about making sure what we want from AI lines up with what it's actually doing.

• Generative AI (GenAI): This AI learns from real stuff like pictures and words to create new stuff, like writing texts or making images.

• Transformers: A smart way to build AI, supercharging it to make stuff. No, not the movie robots.

• Natural Language: That's just how AI folks talk about our everyday language.

• Large Language Model (LLM): Think of it as a giant word map in AI's brain, breaking down language into bits called tokens.

• Token: These are the bite-sized pieces AI uses to understand and map language.

• Training Data: The initial data feast for AI to start understanding things.

• Supervised vs. Unsupervised Training: Supervised is like training wheels for AI, with hints and labels. Unsupervised is letting AI figure it out on its own.

• Generative Pre-trained Transformer (GPT): A specific AI recipe from OpenAI that starts off solo then gets some guidance.

• Compute: AI-speak for the heavy-duty computing power needed to train these brainy models.

• ChatGPT: OpenAI's big hit that lets AI chat, write, and answer stuff.

• Chatbot: A program that can yak back at you. Not a new idea, but getting smarter.

• Turing Test: An old test to see if AI can trick us into thinking it's human.

• Prompt: What you tell AI to do, like in ChatGPT or image-making AI.

• Multi-Modal: AI that can handle different types of stuff – text, images, sounds, and videos.

• Prompt Engineering: Tweaking your request to AI to get just what you want.

• Prompt Injection: Trying to outsmart AI's built-in limits to get it to do naughty things.

• Context Window: AI's short-term memory – bigger the window, more it can remember for a task.

• Hallucination: When AI makes up stuff that sounds right but isn't.

• Deepfake: Fake photos or videos made by AI that look real.

• Frontier Model: The latest and greatest in AI, but also the riskiest.

• Existential Risk: The scary thought that super-smart AI might one day be a big problem for us.

• Explainable AI: AI that can show its work, unlike the mystery-box kind we often see.

READ THE ARTICLE ON AXIOS