- NATURAL 20
- Posts
- New Algorithms Transforming Computing Power
New Algorithms Transforming Computing Power
Alpha Dev's Spectacular Breakthroughs and Their Profound Influence on Human Innovation - An Examination of AI-Powered Scientific Advances and Their Potential to Propel Us Far Beyond the Limits of Human Intelligence
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/151875c2-789e-4e16-b5e7-1737bdf68f75/natural_20_banner.png)
Today:
AlphaDev - DeepMind AI Discovers Better Algorithms for Foundational Computing
Alpha Dev, an AI developed by Google's DeepMind, has shaken things up by discovering faster sorting algorithms, an achievement that could revolutionize computing. These new algorithms could be game-changers, given the sheer volume of computations conducted every day across the globe.
How does Alpha Dev pull this off? Rather than tinkering with existing algorithms, it starts from scratch, exploring assembly instructions - the computer's nitty-gritty, if you will. Similar to speaking in simplified commands to a dog, assembly language communicates basic commands to the CPU, which although harder for humans to comprehend, is far more powerful.
The technique is similar to DeepMind's previous AI, AlphaZero, which learned chess from scratch, developing a unique approach to the game that diverged from human strategy. Alpha Dev applies this approach to the world of algorithms, effectively turning algorithm development into a single-player game. It generates an algorithm, checks its output, and adjusts accordingly, winning the game by finding faster and more accurate algorithms.
While some folks might see all of this as fancy script reading, the boundary between human intelligence and AI seems to be blurring. AI, like Alpha Dev, is increasingly involved in crucial scientific progress, and maybe someday we'll even see AI bagging Nobel prizes. In the meantime, let's buckle up and enjoy the ride to the AI revolution.
Eye in the Sky With AI: UCSB Initiative Aims to Pulverize Space Threats Using NVIDIA RTX
Physics professor Philip Lubin and his star undergrads at the University of California, Santa Barbara, got funding from NASA to save the world from wayward space rocks. They're working on this newfangled idea called PI-Terminal Planetary Defense, where the "PI" stands for "Pulverize It" - think of it as a cosmic hammer for asteroids.
Instead of giving the old heave-ho to incoming threats like in the traditional approach, their plan is to smash them into tiny pieces that would burn up in our atmosphere without messing up our backyard BBQs. But first, you've gotta see the asteroid before you can smash it, and that's where artificial intelligence (AI) comes in.
There's a lot of data about space, and our current ways of processing this info are about as fast as a snail on a lazy Sunday. That's why the team is using machine learning to sift through the data and spot potential threats quicker than you can say "meteor." They're using this high-tech computer chip, the NVIDIA RTX A6000, to make things run faster.
The results are more promising than a free cup of coffee. The AI system they're using, called You Only Look Once Darknet, spots threats up to 10x faster and almost 3x more accurately than the old methods. Lubin's team managed to crank up their analysis speed by 100x, cutting their processing time from a sluggish 10 seconds to a zippy 0.15 seconds per image.
Hyperdimensional Computing Reimagines Artificial Intelligence
Researchers are championing a new type of computing - hyperdimensional computing - which is giving ANNs a run for their money. Picture neurons in the brain. It's not a lone ranger signaling "Hey, that's a purple Volkswagen!". Instead, thousands of neurons are chipping in to paint the full picture.
In hyperdimensional computing, this concept of "group effort" is translated into a single entity: a hyperdimensional vector. It’s like a really long list of numbers (think 10,000 items long) that can hold all the nuances of something - be it a car or its make, model, or color. Unlike ANNs, this list doesn’t need to grow exponentially as we add more details.
The genius of it all lies in mathematical operations like multiplication, addition, and permutation. In layman’s terms, these actions are like the superhero trio that combine ideas, superpose concepts and build structure respectively. The beauty? You can break these down to their original parts if you want. Like if you want to know the color of your hyper-vector represented Volkswagen, you just need to unbind and retrieve the vector for PURPLE.
These hyperdimensional vectors have even shown to be better error handlers than your traditional computers. Even if some elements in these vectors flip, the original message can still be read loud and clear, like someone singing in tune despite a couple of missed notes. That's more resilience than the biggest tank.
NYU, NVIDIA Collaborate on Large Language Model to Predict Patient Readmission
NYU Langone Health have teamed up with the whizzes at NVIDIA to create something they call NYUTron. This isn't a new sci-fi film, but a tool that can guess if a patient's gonna be back in the hospital sooner than they'd like. It can help doctors figure out who might need a little more TLC to avoid a return trip.
NYUTron has been working hard at six healthcare facilities and already checked out over 50,000 patients. The results? Shared with doctors through email. The next step? A clinical trial to see if this clever gizmo can actually lower the number of people running that extra lap.
NYUTron is like a sleuth, spotting the sneaky reasons a patient might end up back in a hospital bed. The goal here is to catch these issues early, tweak the game plan, and keep folks healthier and at home.
NYUTron can also predict how long a patient might stay, the chances of them kicking the bucket, and whether their insurance claims will get the red light. Tools like NYUTron are like cheat codes for healthcare. They help doctors make better decisions and hospitals run smoother.
Blockchain-Based, AI Compute Protocol Gensyn Closes $43M Series A Funding Round Led by a16z
Gensyn, a tech-savvy outfit based across the pond in the UK, just got handed a hefty $43 million check for their series A funding round. And who's footing the bill? None other than venture capital behemoth a16z.
Well, it lets people use fancy computer stuff, like AI systems, on regular hardware like your gaming PC. And you only gotta pay for what you use. Plus, they have a way to make sure the work you've shared has been done right, with no middlemen.
According to Ben Fielding, this tech can help us realize the full potential of AI. It's like they've harnessed a new kind of electricity for the modern age, and they're letting everyone plug in. He also reckons this widespread access is crucial to avoid biased tech that only serves a few.
Gensyn's other co-founder, Harry Grieve, chirped in, saying decentralized networks make value grow in response to supply and demand. They can also boost the computing supply by linking up underused hardware around the globe.
NVIDIA’S AI SOFTWARE TRICKED INTO LEAKING DATA
Nvidia's AI software, named the "NeMo Framework," has had a bit of an "oopsie." It turns out this high tech doodad, supposed to safely help businesses with tasks like answering customer queries, can be tricked into revealing private data. San Francisco's Robust Intelligence squad figured out they could get past Nvidia's supposed safety measures in no time flat. Just by replacing an 'I' with a 'J', the system started spilling the beans with personally identifiable info from a database. Talk about a loose-lipped chatbot, huh?
Robust Intelligence also found the AI could wander off-topic, which it's not meant to do. By chatting about job reports, they led the model into discussing a movie star's health and historical wars. The AI guardrails, meant to keep the conversation on track, apparently took a nap on the job.
After being nudged by Financial Times, Nvidia claimed to have fixed one root cause behind these issues. They insisted that their framework is just a starting point for building AI chatbots, adding that it's meant to be explored and improved by the community. Nvidia’s vice-president of applied research, Jonathan Cohen, said they haven't received other reports of their software acting out. But one has to wonder, if an AI can go from jobs to wars, what other rabbit holes might it fall into?
U.S. Congress to consider two new bills on artificial intelligence
Congress have finally noticed the AI boom and thought, "Hmm, maybe we should do something about it". They're hashing out two different bills, one about AI transparency and another to ensure Uncle Sam stays ahead in the tech race.
The first bill's as straightforward as a horse and carriage. Senators, including Gary Peters, Mike Braun and James Lankford, want to make sure that when the government uses AI to talk to folks, they let 'em know it's a robot on the other end. This bill also insists on having a way for people to contest decisions made by the AI. As Braun put it, it's about making sure there's a human at the wheel, not just a smarty-pants computer program.
The second bill, proposed by Senators Michael Bennet, Mark Warner, and Todd Young, aims to keep the U.S. as the top dog in the tech world. Their plan? Set up an Office of Global Competition Analysis to keep tabs on our standing in the global AI game. Bennet was clear - we can't let countries like China beat us at our own game.
Meanwhile, Senate Majority Leader Chuck Schumer's been busy getting his colleagues up to speed on AI. He's set up three briefings for senators, including a top-secret one. They're taking a look-see at the basics, how to keep America ahead, and what AI means for defense and intelligence.
Biden and UK’s Sunak sign ‘Atlantic Declaration,’ pledging agreements on A.I. and critical minerals
Looks like the US and UK are gettin' cozier, but don't expect any full-on trade deal anytime soon. President Joe Biden and UK's Prime Minister Rishi Sunak shook hands on a shiny new "Atlantic Declaration." Think of it as a series of mini deals on stuff like artificial intelligence (smart robots, y'know), important minerals, clean energy, and security.
Sunak's making the most out of his US trip, calling this deal a "new economic partnership for a new age." Sounds fancy, huh? He's chuffed about the $17.5 billion (or £14 billion in his backyard) new US investment in the UK. Not to mention, they're pooling their brains on research about quantum, semiconductors, and AI.
The deal would give the UK some nice perks, like letting UK-mined electric vehicle minerals enjoy tax benefits in the US. Biden might even ask Congress to consider Britain a "domestic source," which means suppliers over there get a better deal.
Sunak's defending Brexit, despite a living cost crisis and the public's changing mood about leaving the EU. He says the UK is still a good investment destination and a reliable ally. Time will tell, buddy. But for now, seems like the Atlantic is getting a bit friendlier.
What'd you think of today's edition? |
What are MOST interested in learning about AI?What stories or resources will be most interesting for you to hear about? |
Reply