Boston Dynamics Shows Next Frontier in Robotics

Today on the AI Breakdown, we're talking about Boston Dynamics new robot. Before that, on the Brief early reviews of Llama 3 are very, very good. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our YouTube, our Discord and our newsletter. Welcome back to the AI Breakdown Brief. All the AI headline news you need in around five minutes. Today we kick off with a little bit of a catch up on yesterday's main story. Of course, the biggest news of the week in this AI space is Meta's launch of Llama 3. They dropped two different Llama 3 models, the 8B and the 70B, with a 400 billion model promised in a couple of months. While we did anticipate this announcement, given that it had been teased last week, I don't think that most people thought that Meta was going to lean into it so much. It seemed like it might be the type of thing where they perhaps released one or two of these small models as little testers and tastes an amuse bouche if you will, but instead they really leaned into it. Not only did meta announce these llama 3 models, they also built them into an updated overall Meta AI, which features much more prominently in the family of apps including Messenger, WhatsApp and Instagram. There's going to continue to be a lot to unpack with. We barely got into, for example, all of those features that live inside the Meta suite of apps. But what I wanted to do in this brief is instead look at the first impressions of Llama 3 itself. By and large, they are very favorable. Aidan McCloud writes, I'm actually crying right now. Llama 3 is the most incredible model I've ever used. Holy. It's so cheap, so good, has so much personality. It's so fast. Thank you Zuck. I doubted you. I'm so sorry. Now, on the one hand, I would have almost assumed that this was a joke and that it was someone who felt the reverse, but it seems to have been sincere. Perhaps somewhat less emotively, but no less excitedly. Lots of people were showing performance comparisons. A chart that we had recently seen that was all about performance to cost ratio and which featured Mistral models very high in that shows that the Llama 38B chat and the Llama 38B models are also way up in that performance to cost ratio quadrant. And while the cost is higher, they also show that llama 370B and 70B chat have performance that's off the charts relative to other Open source models Maxime Laban writes, Llama 370B looks impressive, but the 8B instruct version is pure madness. It outperforms GPT 3.5, Claude 2 and Mistral Medium. This is on the Arena ELO graph and if you zoom in you see these Llama models getting up into the state of the art territory in ways that are very impressive. And that of course has been another part of the conversation what it means to have this model at open Source. Menlo ventures Dede writes Metislama 3 release is Spotsky's Commoditize the compliment at work cutting edge AI costs at Facebook below 5 billion and making $100 billion OpenAI and $1 trillion Google a commodity overnight. Facebook has a Google Search competitor and owns distribution for a GPT4 class LLM that attracts devs because it's open. Basically the point here is that the fact that Meta has gotten to this level of performance in an open source package pretty fundamentally changes the nature of the game for these other bigger companies. Now of course we still have a lot of market learning to do to see what companies and buyers and developers are going to value, but what there's no doubt about because we've seen it before, is that Meta leading the way on open source is absolutely going to suck a huge amount of talent, especially from a developer perspective, into their orbit. David A. Johnson posted a meme with the AI nerds looking away from OpenAI and anthropic and towards Llama 3 and wrote the importance of Llama 3 can't be understated. The race is to see how good free and open source software models can get. Before the ban in July. Llama 3 unlocks GPT4 level quality, which even if the regime limits models from here, this is a very solid basis on which to build decentralized AI. Purdue PhD candidate Yifei Hu why would anyone still choose GPT 3.5 Turbo API today? It's more expensive than Llama 3 or Mixtral 8x22B and has worse performance. And this again is showing the financial pressure that these new open source models are putting on the big players like OpenAI. Now if you are one of those people who wants to just test this out yourself, Alex Northstar points out that you can try it in Perplexity Playground. This is one of Perplexity's less known features, but is a great way to test open source LLM models on the same prompts. The last thing that I've seen a lot of conversation about came from a section of the Dwarkesh podcast where Zuckerberg was on yesterday. Basically, Zuck Talked about how llama3 was still learning when Meta decided to stop training it, and they were quite sure that it would have gone on training. The only reason they stopped training it is that they felt like they had to make a decision to cut it off somewhere so they could reallocate those GPUs to start testing for llama4. As McKay Wrigley puts it, AI scaling laws are insane. So all in all, a lot of continued excitement around Metislama 3. I anticipate, of course, that by early next week, people who have spent a little bit more time hacking over the weekend with it will start to point out where perhaps it has issues or underwhelms, and I'll try to share that as well for a more rounded perspective. But right now I think it's fair to say that Meta has done it again and that Llama 3 has put the attention firmly back in their orbit. Now, over in Google Land, the company is consolidating its AI efforts in an attempt to focus and improve them. On Thursday, the company said it's consolidating teams that focus on AI across Research and DeepMind divisions. Reuters routes that Google will reallocate its responsible AI teams, which focus on safe AI development from Research to DeepMind, so that they're closer to where AI models are built and scaled. This is not the first time that Google has combined AI units. A year ago or so, it merged its two different research units focused on AI, Google Brain and DeepMind. While that was likely painful internally, in the short term, one of the big challenges that Google has faced in the past, not just in this area but in other areas as well, is having basically competing efforts in different parts of the company, both going after ultimately scarce resources. It only makes sense now that they'd bring that responsible AI team to be closer to the main research team as well. In other words, my view is that this is the next step in a natural consolidation and focusing on, rather than something that indicates any sort of internal turmoil. Speaking of internal turmoil, one company that has seen its fair share of that is stability. The company has lost lots of executives over the last few months, culminating of course with the stepping down of its CEO and board director, Imad Mostaq. Now CNBC reports that the company is laying off about 10%, or around 20 people from its company. That was not the only AI company to have layoffs this week. AI presentation tool Tome also laid off about 20% of its team. One that you know we will be talking more about in the future. The US Air Force has confirmed its first successful AI dogfight. This actually happened last year, but DARPA has now confirmed that an AI controlled jet quote successfully faced a human pilot during an in air dogfight test carried out last year. Of course, as I pointed out lots of times, whatever the conversation about AI is in Washington, the military establishment is not waiting around to see what happens. Lastly, today of course, was the release of Taylor Swift's newest album, a surprise double album, the Tortured Poets Department and Wired pointed out that the leak that happened yesterday shows one of the new challenges that the world is going to face. In short, the problem with AI may not be that people fall for misinformation and disinformation and deepfakes. The problem may be that no one trusts anything to be real. Wired magazine writes, when the Tortured Poets Department leaked, some Taylor Swift fans swore it must be AI. Expect that to be a common refrain. Anyways, interesting things to think about as you listen to the new albums this weekend, but for now that is going to do it for the AI Breakdown brief. Next up, the main AI Breakdown Attention AI Breakdown listeners. Consensus 2024 marks the tenth gathering for all things crypto, blockchain and web3. However importantly, this year's agenda will also dive deep into AI driven transformation and the speaker lineup includes the leading minds and innovators at the forefront of this digital renaissance. Don't miss the Consensus AI Summit to cut through the hype to find where true transformation and opportunity lie. Listeners to this show can get 15% off registration with the code AIBREAKDOWN. Visit consensus2024.coindesk.com to learn more. Some of the folks will be at Consensys this year include Guillaume Verdon, AKA Beth Jesos, founder and CEO of Xtropic, as well as spiritual leader of the Accelerationist movement, Neal Stephenson, co founder of Lamina One and Brendan Eich, the CEO of Brave Software. Again, go to consensus2024.coindesk.com to learn more and get 15% off registration with the code AIBREAKDOWN. Hey friends, one quick note before we get back to the show. If you've been following along, you will know that we launched Super Intelligent last week. Super is a platform to help people learn AI in a way that is fun, fast and very practical. We have hundreds of video tutorials adding dozens each week and we pair them with step by step instructions that get you actually using these tools rather than just reading about them or watching about them. Today for fun, we dropped a set of tutorials related to the big cultural thing, which is of course Taylor Swift's new album, the Tortured Poets Department. If you want to create an AI Taylor sticker or perhaps write a tortured poem, yes, we have the tutorials for you to celebrate this. If you use Code Taylor when you sign up at Be Super AI you will get $15 off your first month. That means your first month on super will just be $5. Check it out again, it's Be Super AI welcome back to the AI Breakdown. Today we're talking about an area that is clearly adjacent to artificial intelligence, which is robotics. If you're a regular listener to this show, you know that we cover the big announcements and updates. We covered the figure 01, for example, which we'll give a little reminder on. But even if you are not a regular listener, you've almost certainly seen a Boston Dynamics robot somewhere on social media at some time. They're the company behind Spot, which is the mobile robot dog, which is already at use in industrial sites, monitoring operations, watching for safety issues, even being a part of some police forces. Well, in addition to Spot, Boston Dynamics has long been working on a humanoid robot called Atlas. Yesterday, the company premiered a new version of Atlas that represents a total break from what they've done in the past. The video shows a robot laying on the ground totally flat and then slowly starting to work its way up. However, it doesn't get up in the way that a normal robot would, or rather a human imitating robot would. Instead, all of its joints seem double jointed and it's able to lift itself up in a totally different and frankly smoother way and then readjust itself to be reconfigured like a human before it walks out of the frame. This video captured a ton of attention yesterday when it was released. Indeed, I'm recording this just about 25 hours after this video came out and it has seven and a half million views, more than 50,000 likes on Twitter Slash X. And that's just the original video on Boston Dynamics first post. That's to say nothing of the thousands and thousands of other posts that used some version of this video. One part of the conversation was simple excitement. Lex Friedman, for example, said congrats to Boston Dynamics on their new electric version of Atlas. Thanks to all the amazing engineering teams at Boston Dynamics, Tesla and others pushing the field of robotics forward. I can't wait to hang out with Atlas and Optimus together at some point. Robot Party but another part of the conversation was about how this changed people's perceptions of How a robot could move. YouTuber Matthew Berman, for example, called Atlas 2.0 a brand new state of the art humanoid robot that moves unlike anything I've ever seen. Dr. Jim Fan from Nvidia said, it took my brain a while to parse what's going on in this video. We are so obsessed with human level robotics that we forget it is just an artificial ceiling. Why don't we make a new species superhuman? From day one, Boston Dynamics has once again reinvented itself gradually, then suddenly. Now this post provoked a lot of conversation. The fact that Jim chose to use the word species and to introduce the concept of superhuman. This is obviously a sensitive one for a lot of people, a lot of folks concerns about the future of AI and robotics has to do with the concerns of designing things that are inherently better than people. Holding aside any of that controversy though, I think the really interesting thing that Jim is pointing out is that designing things to perfectly mirror how human bodies work may actually not be maximizing the opportunities of robotic mobility. Indeed, I think a lot of people's reactions and part of why this video hit so hard yesterday is that it was the first time that they had seen something that actually made it clear that a direct one to one human to robot translation might not be the approach that maximizes the opportunity. Scott Walter at GoingBallistic5 explained a little bit more about what we're seeing in this video. He writes, the Boston Dynamics new Atlas is not double jointed. The knee and elbow both have normal range of motion and cannot hyperextend. The knee linkage mechanism prevents this possibility, but the elbow looks to be a conscious design and not a mechanical restriction. But by having extended range of motion in the hip and shoulder, neck and torso, they can easily mirror the limbs to quickly reverse direction in one step. Looks unnatural, but is a clever solution. However, for many people, what this video really did was reinforce the idea that we are getting very close to these robots being extremely capable with enormous possible implications. Cardano founder Charles Hoskinson said, we are roughly three to five years away from general purpose labor bots entering the workforce. I own a Boston Dynamics dog. Our construction company will likely purchase the successor to this model to augment the workforce. X Prize creator Peter Diamandis said the global market for humanoid robots could exceed 10 billion, maybe tens of billions. The future is amazing and this is a theme that a lot of people have been talking about. Back a year ago, VC Vinod Khosla wrote, the thing nobody talks about is that in 10 years we'll have a million bipedal robots, and in 25 years we'll have a billion. You'll buy yours for 10k and it'll be as important to your life as your smartphone is now. The fascinating thing about this, and why VCs get so excited about it, is that this really is a space that has a virtually uncapped upside. We have no idea what the total addressable market for robotics will be, but it's going to be enormous. So big, in fact, that it's very likely that there will be multiple big multibillion dollar companies that are created in the space. Two of the others that are worth keeping an eye on are figure, which recently closed a $675 million funding round with investors that included Microsoft, OpenAI, Nvidia and Jeff Bezos, and which you might remember premiered this video of their Zero1 humanoid robot, powered by OpenAI's ChatGPT to interact with the human in a very natural way. When the person interacting with the robot says that they're hungry, it's able to identify an apple from in front of the robot and give it to the human. Figure CEO Brad Adcock was just in Forbes as well, discussing the company's ambitions. Forbes summed it up on Twitter by saying, brett Adcock wants his company's humanoid robots to fill job shortages in factories and do work that is unsafe for humans, and someday take out your trash and make you coffee. And this is really the optimistic side of this. Again, Vinod Khosla from last November Bipedal robots have the capacity to transform every vertical, from elder care to factories and farms. Few are preparing for how this will radically change gdp, productivity and human happiness. These robots could create enough value to support the people they replace. In a further prediction, he wrote, In 25 years there could be a billion bipedal robots. A million in 10 years, doing a wide range of tasks, including fine manipulation. We could free humans from the slavery of the bottom 50% of really undesirable jobs like assembly line and farm workers. This could be a larger industry than the auto industry. Now, of course, the way that society adapts to 50% of those undesirable jobs being gone remains an open question, although one perhaps a little bit outside of the scope of this particular episode. The last note for those of you who are using this as a chance to get up on your robotics, another big player in the space is of course Tesla. Their humanoid robot is called Optimus, and frankly we haven't seen any big updates from it lately. I think given the figure 01 as well as now this excitement around the new Atlas. I would be very surprised if some video didn't drop soon. For now, really crazy and interesting times. Like I said, an offshoot and related area to AI. But that's going to do it for this AI breakdown. Until next time, peace.