Q* why AI that is "Good at Math" a "Threat to Humanity"? | Q star

Today:

Q* why AI that is "Good at Math" a "Threat to Humanity"? | Q star

This leaked document talks about "Q-STAR" and suggests it's some kind of super advanced AI. Some smart folks in the AI field, like Dr. Jim Fan, are trying to figure out what "Q" could be. They think it might be a mix of different AI tech, like what Google used for their AlphaGo program and some other fancy AI techniques.

The document that got everyone talking is pretty well-written, like an insider's scoop. It talks about Q doing some mind-blowing stuff, like breaking advanced encryption and showing signs of what they call "metacognition" – basically, thinking about its own thinking.

WATCH THE VIDEO ON YOUTUBE

Amazon’s Q AI assistant lets users ask questions about their company’s data

Amazon's cloud wing, AWS, rolled out a new chat tool called Amazon Q. It's like a smart helper for businesses to dig into their own data. It's like having an AI buddy in your pocket, ready to answer all sorts of questions about your company's data. You can use it to dig up company policies or make sense of complex code without having to rummage through a bunch of documents.

You can get to Amazon Q through AWS's main control spot, your company's own doc pages, or even in places like Slack and other apps. The AWS head, Adam Selipsky, made it clear that whatever you ask Amazon Q won’t be used to make their AI smarter.

Amazon Q can mix and match with different AI models from Amazon Bedrock, which is like a library of AI brains. This includes stuff from Meta and Anthropic. Businesses pick the brain that suits them, hook it up to their data, and set up Amazon Q.

For now, it's only for Amazon Connect users, which is for folks running call centers. But soon, it'll be in other Amazon services, like supply chain and business intelligence tools. Each Amazon Q will look different depending on where it's used. In Amazon Connect, it's like a silent partner on customer calls, giving agents quick answers without them searching.

READ THE ARTICLE ON THE VERGE

Amazon unveils new chips for training and running AI models

Amazon just rolled out some cool new tech chips, Graviton4 and Trainium2, whipped up by their AWS cloud folks. These chips are top-notch, boasting better performance and energy use, especially for stuff like machine learning and AI. 

Amazon's not sitting around, though. They're whipping up their own chips for AI stuff, like making and running those smart models. First up, they've got this beast called AWS Trainium2. It's a monster, like four times better and twice as energy-smart than the first one they made back in 2020. You can grab these in clusters of 16, or go wild with up to 100,000 chips in this UltraCluster thing they got. We're talking major power, like 65 exaflops in total, which is tech talk for "really, really fast."

This Trainium2 can train a huge AI model in just weeks, not months. We're talking bigger than OpenAI's GPT-3, and that's saying something.

Then there's Graviton4, Amazon's new kid on the block for running these AI models. It's an upgrade from their previous chips, with more power, more cores, and more memory. Plus, it's got this encryption thing for better security, which is a big deal if you're into that.

READ THE ARTICLE ON TECHCRUNCH

NVIDIA BioNeMo Enables Generative AI for Drug Discovery on AWS

NVIDIA's got this cool tech called BioNeMo that's all about using AI to find new drugs faster. They're teaming up with Amazon Web Services (AWS) to make it easier for folks in the healthcare and drug research biz to use this tech.

This big announcement at the AWS re:Invent event means that if you're working in healthcare or life sciences and you're using AWS, you can now tap into NVIDIA's BioNeMo. It's like a turbo boost for drug discovery, using AI to quickly figure out stuff like what proteins do and which drug candidates might work best.

The cool part about BioNeMo is that it's not just some off-the-shelf thing. Drug researchers can tweak it with their own data to make it even better. They've got this model called ESM-2 that's really good at predicting protein structures, and it works super fast on NVIDIA's hardware.

Plus, NVIDIA and AWS are making it easy to use these tools. Whether you're managing your own setup with AWS ParallelCluster or using NVIDIA's managed services, you've got options to build and run AI models for drug research.

READ THE ARTICLE ON NVIDIA

Introducing SDXL Turbo: A Real-Time Text-to-Image Generation Model

Stability AI just rolled out this new thing called SDXL Turbo. It's a way to make pictures from text, super fast. They're using this cool trick they call Adversarial Diffusion Distillation (ADD), and it basically lets the model whip up images in one shot, real quick, while still looking sharp. They're saying this is mainly for folks doing research or just playing around for now, not for selling stuff.

SDXL Turbo takes what the older version, SDXL 1.0, did and makes it even better. Thanks to this ADD, it can make images in one step, kind of like how those AI art generators (GANs) work, but it doesn't have the usual problems like fuzzy or weird-looking images.

When they put SDXL Turbo to the test, they checked it against a bunch of other models by seeing which one could best turn a piece of text into a picture. They had people look at two images and pick the one that matched the text best. SDXL Turbo came out on top, doing a job that would usually take a bunch of steps in way fewer steps, and without losing any of the picture quality. That's a big win because it means it can make great images without needing a ton of computer power.

READ THE ARTICLE ON STABILITY AI

Dell Inks $150 Million Hardware Deal With AI Startup Imbue

Dell just scored a big win, landing a $150 million deal with a hot AI startup called Imbue. This is a big deal for Dell since they're trying to make a name for themselves in the AI world. Imbue, which is kind of a new kid on the block in AI, recently got $200 million from investors, including Nvidia. These guys are building their own AI from the ground up, which is no small feat - it needs a ton of computer power.

Matt Baker from Dell is pumped about this. He's saying this AI boom is as big a deal as when the internet first showed up. The twist here is, instead of doing the usual cloud computing rental like Amazon and Google do, Imbue bought the hardware straight up from Dell. Josh Albrecht, the tech head at Imbue, says this way they're not tied down to one big tech company and can keep their independence.

Dell's showing the AI world that startups don't always have to go to the big cloud providers. They're giving Imbue a mix of server setups - some for quick tests and others for the heavy-duty AI building. Imbue's system, managed by Voltage Park, is designed to help them build really smart AI tools that can do things like check code for bugs or even plan vacations without needing a human to hold their hand.

READ THE ARTICLE ON YAHOO! FINANCE

Google, Symphony unveil AI JV to help banks manage voice calls compliance risk

Symphony, a tech firm, and Google are joining forces to help banks and investment firms keep an eye on voice calls. Why? Because the U.S. government's been cracking down hard, fining companies billions for not keeping tabs on business texts and messages, especially during the COVID lockdowns.

Now, the heat's on for voice and video calls. Lots of these aren't being recorded, and that's a big no-no for the regulators. Symphony's using Google Cloud's smart AI tech to beef up its Cloud9 voice tool. This means better voice-to-text and understanding what's being said in those calls (that's the natural language processing bit).

Cloud9 is used by big-shot banks for their trading teams. This upgrade means they can write down what's said in calls better and spot any sketchy talk.

READ THE ARTICLE ON REUTERS

Valley Is Leveraging Contextual Generative AI For The Sales Industry

Valley, this hot new startup, is shaking things up in the sales game with some fancy AI tech. The brain behind this is Zayd Ali, a 21-year-old whiz kid who's already sold a couple of companies. Valley's big idea? Use AI to make setting up business meetings a breeze, something that used to be a real headache.

So, how's Valley doing it differently? Instead of just throwing a bunch of data at their AI and hoping for the best, they're focusing on quality. They're training their AI on the best sales tactics out there, making it smarter and more efficient with each use. This means the more you use it, the better it gets at understanding your needs.

Now, Valley isn't just about big ideas. They've got some serious brains on their team, with folks from places like Samsung AI and Columbia University. They're building on existing AI tech but fine-tuning it to nail appointment setting. And they're not just grabbing any data they can get their hands on; they're picking the cream of the crop, making sure each piece is just right for the job.

READ THE ARTICLE ON FORBES

Pika Launches AI Video Editing App And Announces $55 Million In Funding

Last winter, Demi Guo and her brainy Stanford pals tried to make a movie with some fancy AI for a contest by Runway, a hot AI video editor company. They thought they'd nail it, but no dice – they didn't even get a nod. Turns out, making a slick film with the AI tools they had, like Runway and Adobe’s Photoshop, was a real headache.

Frustrated but not beat, Guo and her buddy Chenlin Meng, ditched Stanford to kick off their own gig, Pika, aiming to make AI video making easier. Fast forward, and over half a million folks have played with their software, making millions of videos weekly. This buzz caught the eye of big-money folks in Silicon Valley, leading Pika to bag $55 million through three quick funding rounds. The latest round, a cool $35 million, pumped Pika's value up to somewhere between $200 and $300 million.

Pika's tool, originally just on Discord, let users type in stuff like “a robot walking on the beach at sunset” and get a video back. Now, they've rolled out a web app where you can tweak these videos, like slapping sunglasses on that robot.

Pika first toyed with just making anime, thinking realistic videos were out of their league. But they surprised everyone, including themselves, by pulling it off fast. They even added features overnight when asked. Pika's now leasing a bunch of GPUs, some from Friedman's cluster, to beef up their AI model. They're also figuring out how to dodge legal headaches with copyrighted stuff. The plan next year is to grow the team, focusing mostly on engineers and researchers. 

READ THE ARTICLE ON FORBES

Meta AI develops a non-invasive method to decode speech from brain activity

Meta AI's brainiacs have cooked up a slick way to figure out what you're saying just by peeping your brain waves. This is huge for folks who can't talk due to strokes or illnesses. Before, the only way to do something like this was brain surgery with gadgets attached directly to your noggin. But this new method? It's chill. No surgery needed.

They use this fancy thing called magneto-encephalography, which snaps thousands of brain activity pics every second. Then, they trained an AI to translate these images into speech. It's like a one-two punch: one part of the AI digs into your brain activity, and the other figures out what you're trying to say.

They tested it with 175 people listening to stories and sentences. The results? Pretty neat. They got it right 41% of the time on average, and sometimes even 80%. If the AI goofed, it usually guessed something close to what was meant.

READ THE ARTICLE ON MEDICAL XPRESS

Critical Vulnerability Found in Ray AI Framework 

Ray, this tool used for AI stuff, has a big security hole. Cyber experts at Bishop Fox say this flaw lets bad guys get into all parts of the system. They can mess with jobs, steal info, and run their own code.

This particular bug, tagged CVE-2023-48023, is a real head-scratcher. Ray's set up in a way where it doesn't check who's coming in or what they're doing. Even if you try to lock it down with some techy stuff, it's still pretty much a free-for-all.

To make matters worse, this isn't the only problem Ray's got. Another group, Protect AI, found some more bugs. They told Ray's bosses at Anyscale, but get this: Anyscale just shrugged it off, saying it's all part of the plan. Sounds fishy, right?

Bishop Fox also spotted some other serious issues with Ray, like a bug where someone could trick the server (that's CVE-2023-48022) and another one where Ray's not too picky about what it accepts (CVE-2023-6021). These are big deals, but Anyscale's acting like it's no biggie.

READ THE ARTICLE ON SECURITY WEEK

OpenAI rival Hugging Face says it’s seeing more client interest after Sam Altman fiasco

Hugging Face, another big player in the AI game, is getting more attention from companies since things got messy at OpenAI. Their CEO, Clément Delangue, thinks businesses are nervous about putting all their eggs in one basket with just one AI provider. It's not just Hugging Face either; Cohere, another AI heavyweight from Toronto, is also getting more calls.

This all started when Sam Altman, the head honcho at OpenAI, got kicked out but then came back real quick after some drama with the staff and money folks. Cohere sent a message to their investors saying they're all on the same page, unlike what's going down at OpenAI.

Cohere's trying to play it cool, saying companies want stable, drama-free solutions. Meanwhile, Hugging Face, based in New York, bagged $235 million this August, hitting a whopping $4.5 billion value. They're backed by some heavy hitters like Google, Amazon, and Nvidia.

Other companies in the AI race, like Anthropic and Google's Bard team, are keeping mum on whether they've seen more interest since the OpenAI shakeup. And there's a bit of a warning: if the U.S. doesn't get its act together on AI, China might take the lead.

READ THE ARTICLE ON CNBC

Anna Indiana is the world's first all-AI singer-songwriter. She's deeply mediocre.

Anna Indiana, the world's first all-AI singer-songwriter, has stirred up a storm online. She's a digital creation, cooked up with some fancy AI tools like ChatGPT 4 and Musicfy. Her debut track, "Betrayed by this Town," is all about heartbreak and despair.

The creators behind Anna haven't stepped into the limelight yet, but they've used a mix of tech to bring her to life, including some open-source Python magic for the music and ChatGPT 4 for the lyrics. Despite all this, her first music video got a ton of flak online, with folks ranging from YouTube critics to comedians dissing it.

This isn't the first time we've seen virtual musicians. Remember Hatsune Miku or Miquela? But Anna's different because she's supposedly 100% AI-made. And let's just say, her music won't be winning any awards for creativity. It's kind of eerie, too – her name itself is a nod to a future where AI runs the show.

READ THE ARTICLE ON BUSINESS INSIDER