Robots Are Having Their ChatGPT Moment. Here's Proof

While America dominates the AI race, our robotics companies have been playing catch up with China. But we got new manufacturing plants from both Figure Robotics and One X. They're opening here in the good ol' U.S. of A? And the tide might be turning. Hey, hey Apache Helicopters flying to Kid Rocks. We'll let you know why robotics and physical AI is the next big thing and why this robot grabbing a raspberry might be the coolest thing you're seeing this week. I hope it's not the coolest thing you're seeing all week. Oh don't worry there's better things getting ready for my poop video. Oh great. Also what's the deal with goblins? We're gonna have to explain why OpenAI dropped a full-on blog post about strange creatures and why they keep showing up in your chat GPT. And in our inaugural Sloppies AI Video Award nominees we have Amazon Audio Today our AI generated shopping show is exploring the Welmedix Rapid Relief Diaper Rash Cream. This uh this is not a sophisticated work environment. No Kevin this is AI for Humans. Welcome everybody to AI for Humans your twice a week guide to the wonderful world of AI and today robotics because those two things are tied in like peas in a pod. What's a good robot and AI metaphor Kevin? We have to think like metal meeting brains meeting... What happened there? Did I freeze or did you freeze? Nope I was just letting you go. I was just letting you I was letting you get there Kevin you were taking the walk. This is perfect. I'm in an Airbnb this is a weird place. Fire up your Neuralink buddy get those get those ones and zeros. Let's talk about American robotics. So we've been covering this story for a while. Robots are coming they are the next big thing. When people talk about physical AI what they mean is the idea of AI that is moving into the physical world. This is driverless cars this is coffee pots that talk back to you which I'm very excited for and this is humanoid robotics and Kevin China has been crushing this world for a while but we got some very exciting news this week in the world of American robotics and maybe just maybe this is where we see the tides start to turn in terms of America coming back a little bit. Should we talk about some of the things that are popping up right now? No because I'm still thinking about a bipedal coffee pot like that would be like imagine like it's the best part of waking up is actually catching your cup like it would wake you up before it would jump on you the caffeine. No I'm saying it's running away from you. Oh so I see it's kind of like that robot alarm clock that used to spin away from you. Yes there we go like I just need my cup of coffee oh come on come back and it's spilling hot coffee all over your living room. I would destroy that thing so fast I would smash it and the next day I'd be like just give me Mr. Coffee again I don't care. By the way Mr. Coffee if you're out there I'm more than happy to work on you with this exact idea so get in touch we are looking for a new sponsor. Let's talk about let's talk about American AI. So Kevin there's a couple big videos that came out there are videos but they are videos showing off physical production plants for two American companies 1x Robotics and Figure Robotics are both now having manufacturing lines of robotics. Now there's some pretty big news out of like how many they're going to make per year. Figure says they're going to make one an hour that's what they're making now in fact they had a kind of a cool stat if you watch their video it's very inspiring you see a new kind of industrial line that we haven't seen new American factories like this for a while but they were making one per hour used to be one per day so they're moving up. 1x has said they're going to make 10,000 per year. Now I want to hear your thoughts on this because you and I have been covering robotics here for a while. China obviously has a very large lead we just saw the half marathon where a Chinese robot beat a human time in the half marathon. I want to hear your thoughts on this and I'm very excited to talk about EXO which is another American robotics company but first let's talk about all this manufacturing that's coming back to America. Yeah I mean great we need it well we've needed it sorely I'm sure that there is a room full of school children in Shenzhen or in one of the outside regions giggling about this amount of productivity because they've got some 3d printers and they're churning out 10x this capacity but that's not to shake like look we're we are where we are we outsourced and shipped off a lot of our capabilities over the last few decades and we're trying to bring that back and so I think that's great there's been a little bit of skepticism about some of the figure numbers I saw like people saying like look yeah you could make x robots per hour but that's assuming you're running 24 7 are you running 24 7 365 are things changing there then again I still think that the numbers should be celebrated right yeah you pointed it out they went from the one robot a day to one robot an hour how much lower can that go as they optimize their lines and their throughput I mean amazing we need it let's go I think that actually that's an important thing for everybody to understand here kind of is like this idea of what physical AI is and this is the thing that like Jensen Wong and all these people have been talking about for a while is this idea that all the AI models that we talk about right now are kind of locked in our computers or locked in the cloud right and they're very smart they're getting smarter all the time but physical AI is taking those models and bringing them out into the real world and in some ways it's been successful so far and in other ways it's still in its very baby steps I think this is a good transition to talk about what wired magazine just recently called that they think it's will night he said I've covered robots for years this one is different he believed ectus robotic claw feels like we are approaching a chat GPT moment for the physical world so Kevin I saw this video this morning um and the most impressive thing was most of the time these robot videos say like 1 10th or or like slowed down they're like way slowed sorry they're sure sped up no it's like 5x speed yeah it's like hey look at look at how fast this thing is moving when we speed it up a hundred percent exactly and in this video of this robot grabbing a raspberry which we were probably going to show in our video version a thousand times here so just get ready this video is slowed down it's slowed down quite a bit because when you watch it full time it's really impressive so I want to hear your thoughts on EXA and it's another American grips a raspberry like me at a farmer's market before they can tell me that there's no free samples well that's better than I thought where you're gonna go I thought you were gonna say something else but I'm happy keep going here what like chocolate before my dog gets to it in the kitchen gallon is that what you're gonna say that's exactly what I was thinking what did you have on deck but oh I had nothing on my mind just keep going please um no so like the yeah the fact that they slow it down amazing the fact that this thing was trained entirely in simulation yes to me is is one of the coolest things about this so they've just got uh they basically have a a highly detailed physical simulation of the world and they let this robot arm go and experiment and learn how to grip different things I love it trying to grip the um like the headphones yeah it's kind of struggling with the spaghetti wire a little bit like that makes me giggle but the notion that it can simulate this stuff uh you know thousands and millions of times over and then apply that to the real world and not have it uh completely destroy the raspberry as it goes to grip is just like I bet that to me that tickles me well and I think that's the important thing for people in our audience to realize if you haven't been paying attention as much to the robotic side of AI is that just like with AI models which learn over time get better over time with big bigger parameter models they get better results often they're finding this with with real world physical AI too right because what you said is the simulation uh training right so they're training in these simulated environments if you remember like I think it was not last year's NVIDIA conference but the one before there's that video of all those robots that are in the kind of like square room when they're walking around falling down all of that stuff is happening all the time in these kind of robotic server brains and they can push that directly into each one of these things but until they get into the physical world they're not exactly sure how it's going to work so what's cool about seeing these EXA videos is that clearly we are now processing through to the next level of what this capability is and I think one of the things I find really fascinating about this and I'm curious to hear your thoughts on this is about the speed at which this will grow because I think if three years ago you had told somebody chat GPT would now be a governmental uh you know essentially a huge government risk and and maybe endanger the jobs of like you know millions and millions of people they would have said you are an idiot and they did to us Kevin they said that multiple times to us back then they still haven't stopped we might have been right about a couple things but they yes but looking at this stuff now I'm curious to think like project out two years five years in the future for robotics AI and kind of where do you see this stuff going I mean one of the things that's much harder about physical AI is manufacturing and the real world and the safety elements of it like oh there's a great creepy video that I saw from a Chinese company called um uh Chi Robotics that you have to watch here uh while we're doing this but it's a video of a little child who's in a like a nice house and there's this very charming faceless kind of buff robot kind of helping it out and it makes me like when I see that I'm like well are they sure that safety of that robot is strong so anyway that's a long setup for I'm just curious to hear you know kind of what your thoughts on like what we'll see in the next two to five years well that's it's like look the the edge cases on these things can get deadly within a millisecond and there's just no way around that like I'm I know that we there will be certain safety systems implemented but we're already seeing these early robots where they trip on something that's unexpected or um you know one variable gets shipped in the wrong way and it goes full speed one you know servo flips and like a robot arm or leg or torso spins around with enough force that it could uh totally brain you yeah and you know like that's that's it's it's funny in this isolated in my twitter timeline as I scroll sort of thing but if that's in the living room and I'm uh you know I'm away and it's trying to it's trying to do dishes and my dog is around like what is going to happen I bet a mishap could get very bad very quickly I have a worse idea what if the robot sees your dog screwing up your place and like becomes buddies with your dog and then they together destroy your house you could come home to much worse destruction so that this if a robot is mounting the pillows from my couch I'm going to be very upset I that should not have been in the simulated environment well how dare this will do not make that video we don't want to see it don't make it go ahead well let's see that video what what fascinates me is the idea like the the machines making the machine we've already seen there's certain factories in China where this exists today but I you know we're going to gun here as well like once these robots get capable and dexterous enough to build themselves the exponential there becomes very wild and for home health care and elder care for doing the dishes and the chores um for uh you know dangerous um you know trade jobs all that stuff is great the the elephant in the room obviously is military applications right now like these robots just have to be good enough to hold a a nerf gun uh and be able to be made in mass so it's like there's there is a war for uh a very literal and figurative war happening and um we do need manufacturing capability to get there um it just blows my mind to see how far we've come so quickly but yeah in in five years time Gavin are you at a magnolia hi-fi browsing robots uh for one to take home or is that 10 or 15 away I think in five years time a robot is browsing me and saying I want to take that person home and be in the robot just for novelty you're at a vintage shop exactly I'm in the humans please help us brush up I do think in five years I do want everybody to has a just a kind of a point to think about remember what chat gpt2 or 3 look like gpt2 or 3 if you were in this space before chat gpt and just how stilted and weird it felt gpt2 especially that was like you know five years ago so just think about from where we were then to now is a big deal although Kevin I did send you another video here this is a video where it's going to take a while right if you watch this video of this unitary robot at a big uh it looks like a presentation in China where a lot of people are in their outfits uh this reunity robot supposedly has struggled with the heat in the room and it's just not doing very well as it tries to keep up no this is me at the reception uh five beverages in trying to prove to my wife that I still got it and she's she's doing her best to get me back to the hotel but I swear no no once they play Baja man it's over I let these dogs out you listen the only thing that can contain a flailing robot like this is a cool airpods-esque case and the neocarrying case that they're showing up yep Gavin is folding himself put me put me in there put me a coach you know I'm surprised this isn't like a Japanese sleeping pod sort of trend right now is to just put yourself into the carrying case of shame um but it is really cool to think about like how these things are going to arrive or be delivered or checked onto a flight as we bring our robot servants with us oh yeah can you imagine a carry-on with that like put the carry-on in the in the space and you're like what my robot doesn't fit what am I gonna do man uh we should also chat about something really quickly that you mentioned was this idea of government and defense and there is a big story that's kind of breaking right now around the trump administration and the mythos model from anthropics which we know was kind of a point of contention well Kevin there's been some flip-flop here and there's a couple different versions of this story but according to the wall street journal right now the wall the white house is saying they are pushing back against the further rollout of the mythos model to more companies because they want more of the compute for themselves so this is a really fascinating thing that there's a little bit of egg on face maybe on the on the administration side because the anthropic model clearly is being used by a lot of people both for defense and for government use cases i think this is going to get weirder before it gets less weird like i really do think we're we're kind of with all we just talked about what the possibility of robots mean for both our government our industry our economy and the world economy we are really moving towards some sort of nationalization of ai and i don't think we have to spend that much time on this because it is a very deep subject i've been thinking about trying to do a deeper video about this because it does feel like something that's coming sooner rather than later but again mythos even though we don't have access to it is making headlines behind the scenes so this model does feel like it's something pretty big well it is you know if it were generating um shakespearean prose at a level unseen before i don't know that so many nations would be or you know the government would be batting an eyelash it's because it's specifically targeting cyber and you know to that end open ai doesn't want to be left out so sam altman is you know the meme he's they don't even know i do cyber as well he's arriving to the party while everybody else is dancing with mythos um there's a new 5.5 model according to sam altman i said we're starting roll out of 5.5 cyber a frontier cyber security model uh to critical cyber defenders in the next few days so he's he's getting there he's there too he's there too don't count him out we are going to talk more about the goblins in just a second but i do want to shout out one thing from grok because as opening eyes work on cyber security as anthropic is working on mythos and like whatever they're doing helping the government without us being able to use it grok has created an imagine agent kevin so that you can now make your videos in peace you can sit there with the grok ai and make something so this is basically grok imagine has now integrated almost like a comfy ui style interface you see a lot of these ai interfaces but i do want to shout out very quickly i went and tried this and i used my compute for the best possible use case i created a video of a jack buff body guy who landed on an abandoned beach found some uh found some protein powder and then did this yes i found it come on out you come got it beach games are back on beach games baby oh wow oh wow is right okay here oh boy this is i mean it's like you know i had plenty of guys that used to i knew i rolled with a lot of 16 year olds gavin that would beater cars like real real beater cars held on with duct tape but then they would go and put like you know thousands of dollars of uh afterglow street effects or subwoofers in the trunk that would rattle the rusted uh lids just that's this guy it feels like they bolted on like some powerful agentic cools to a video model that is just from you know well so that's 2021 well that's what i want to say like i but by the way grok imagine doing very well in many different ways but like this is a tool that supposedly allows you to kind of like create a video with an agentic tool and it has agents in it it spins up agents it wasn't great i'll share some of my uh kind of like generations here i had to kind of stitch those videos together it actually said it called ffmpg but then it couldn't figure out how to do it so like there's this element of grok imagine right now where it's very powerful in some ways but this tool feels a little underbaked i do want to shout out our friends at glyph because glyph is actually a company that's put together a lot of really good ai agent tools that do the same sort of thing and just know that like ai agent video tools are coming in a pretty big way and this is just another example that grok's putting their attention towards it but i think it's a really interesting thing at least yeah look for those that just got the audio only in this video um the the main character uh mr gains is it you know slams the protein powder into the sand and then unburies it and then in the next scene he is doing two-handed pull-ups yes well the third hand holds his his uh protein drink well in the first one the shaker bottle is pouring up into his mouth while he's doing chin-ups and then yes there's a third hand coming out of his armpit uh very old spicy while he's doing some pull-ups and so that's when i go like look it's cool that they are you know iterating and giving you different tools to make these videos but the core physics and world modeling and understanding of the underlying video model just feels so dated to yeah i mean you know it's it's fun it's fun and it's the beginning stage of something you know last week we mentioned this idea that claude and blender can now work together and actually kevin for this week's sponsor i did some work with that i'm about to do something i never thought was possible on a laptop computer vibe code a playable 3d prototype while running blender a live preview and have 40 open tabs without crashing a single thing we're so thankful to hp and intel for sponsoring ai for humans this week and sending us this absolute beast of a pc i've got cloud code running on this hp zbook fury and i'm gonna make something special that i've been dreaming about for a while a 3d game about raccoon gangsters here's the thing about vibe coding cloud code is doing the heavy lifting on the ai side but you also need a machine that can run the stuff you make right now i'm running multiple instances of cloud code in the terminal it's refreshing a live preview every time i push new code blender is open for the 3d assets and i've got a bunch of other tabs open that i'm just never going to close and on most laptops that is where everything comes to an absolute stop but the hp fury has an intel core v9 pro processor which means that i'm not choosing between running my coding agent and doing everything else and with the 256 gigabytes of ram this computer has nothing is fighting for resources when you're working with 3d rendering and real-time previews that's where the nvidia rtx pro 5000 blackwell gpu comes into play i'm previewing the game as claude builds it every time i ship a new feature i'll check it out and the most fun part is that it's so much faster than any laptop i've personally ever used cloud writes the code i iterate it pushes the code we get a new 3d model some of them are better than others but still i got a version of raccoon gangsters in 3d running on my laptop check out the link in our description if you want to spec out your own zbook fury and thanks again to hp and intel for sponsoring ai for humans everybody make sure to like and subscribe to our channel as well we love you you love us we love you when you love us and we do not get paid to do this channel so please like and subscribe if you're watching this video we really do appreciate that i mean we did just technically get paid because we had a sponsor like we don't the point is it all good one here it is one dollar goes in two dollars go out that's the way we operate baby so it's always at a loss always be losing abl that's our business models always at a loss all right kevin we have to talk about goblins and i never thought i would say that on this podcast i'm very excited to say on this podcast but yeah there is a share your sketches gavin people don't know this for decades gavin if he does a goblin a day he i have a goblin on my bag right now in fact um uh ai is infested with goblins and this is a very interesting story actually underneath on the surface it looks like a weird thing you'd be like what the hell is opening i talking about goblins to set the table here opening eyes come out and said yes we have a mentions goblins too much and this came from a post about a week ago where somebody was looking into the codex um instructions custom instructions and saw something say please don't mention goblins as well as a bunch of other things and what they found there was literally a prompt inside of codex that said quote never talk about goblins gremlins raccoons trolls ogres pigeons or unless it is absolutely and unambiguously relevant to the user's query they are wasting tokens and context on every prompt to specifically tell their world-class foundational model stop talking about goblins and what's so interesting about this is you think okay well this is kind of a stupid story like yeah sure stop talking about goblins but what was interesting about this is it really gets to the heart of how some of these models are trained because this appeared first in gpt 5.1 this is very funny to me but it's because of the nerdy setting if you remember in that gpt in 5.1 they allowed you to choose from different personalities and in the nerdy setting goblins would be mentioned every once in a while and then it started to see interactions and different things and then when each further model trained because these models kind of train a little bit on each other the goblins kind of expanded and they started becoming a bigger part of the reactions and the responses to things so this is what happens when you have a model that's kind of training across the way now sam altman has gone out and said like look we understand this is a thing it's kind of a goofy thing you know there's a i have a conspiracy theory a little bit that like sam is making this goblins into a much bigger thing to like distract from maybe some of the other stuff going on or opening eye but they did say that the next like gpt6 the next large huge model will not mention goblins as much so you can feel safe that you no longer have to worry about goblins appearing in your thing although i will say to me i kind of think this is fun why not make ai fun maybe this is like where we get to like ai being magic and suddenly goblins are everywhere i mean bring forth orcs and everybody else from the lord of the rings universe why not dobby well that's i mean that's the funnier take on this is not that they're distracting from like lawsuits or government partnerships or that they're bleeding out from uh you know people switching to anthropic it's way more fun to believe like their model got so powerful it found the hidden connections and the goblins are everywhere gavin i cannot wait to dig further into that conspiracy theory goodbye weekend sorry why we will not be spending the rest of the time in there all right kevin i can't wait for this this is very exciting i've been waiting to do a version of this for a while we are going to start a first ever ai for humans award show nominee by nominee this may take us multiple years before this award show happens but it is the ai sloppies and the ai sloppies awards okay kevin the first sloppy award uh nominee of ever it is wait is it hold on just so i understand the context of this bit yes this is a nominee or they're winning a sloppy or and no no no they're not they can't win yet this is what's the judge panel for these sloppies who is okay i think it's probably just you and i maybe just me we'll have see what your attention span is for this and my attention span is for this i'm already i'm already in the weeds about the sloppies itself and not the nominee you're in you're in okay so today's nominee this is the first nominee we'll have many more to come today's nominee is the amazon ai podcast feature and this is something i didn't know existed uh katie netopolis who's a very funny um poster on x but also other places writes for business insider he's right for buzzfeed brought this into my world and what you can do now on amazon for some particular um products is there's a little button that will show up and it basically gives you a google like ai podcast about that product and let's play katie's which was about diaper rash so let's hear this one diaper rash cream today our ai generated shopping show is exploring the well medics rapid relief diaper rash cream by the way what an unfortunate pronunciation of well medics well medics well my dicks needs this uh rapid relief cream it's getting better this sloppy is doing very well okay i'm playing that again ai generated shopping show is exploring the well medics okay okay keep going diaper rash cream emma what makes this hospital grade cream different from standard diaper rash products well it's really interesting this cream uses a dual action approach just show i mean because i know amazon's like they said huge shout out to the aws bedrock ai teams over there but they're doing huge work they really are but it just shows how great those notebook lm voices are like definitely trained to be far more conversational but okay i'm sorry let's get back to the well medics cream instead of just zinc oxide it combines that with white petrolatum to create two protective barriers this helps both treat existing irritation and prevent new issues from developing okay so let's pause this here so it's fascinating it is true it's fascinating i wanted to go and see is this is this what our podcast sounds like to our wives yeah it probably does i think it probably works to our wives okay i wanted to go and see what other things could do this and like there's a few you've go and find and it wouldn't there but like kevin i was like okay what would be the do the dumbest weirdest thing to hear a podcast on and without telling our listeners i just want to play this clip that i created and i want everybody to listen to what ai can do now today our ai generated shopping show is exploring the vaulted skulls realistic dog poop prank jake what makes this particular gag gift stand out well it's all about the attention to detail in the design at four inches long it's sized perfectly for believability the chunky texture and authentic brown coloring make it a real showstopper and i mean that quite literally when people spot it oh my oh my how are people typically using this um unique item yeah great please you know it's become a go-to for office pranks and party gags the material is really durable so pranksters can use it repeatedly according to customer feedback the realistic appearance is a huge hit though some what who mentions that keep going keep going all right okay yeah i suppose that's good what makes this different from other prank items versatility really sets it apart unlike whoopee cushions or fake spiders that are obviously gags this can be used in so many creative ways from halloween decorations to theatrical props it's super easy to clean with just a damp towel who would you say this isn't right for you know if someone prefers more sophisticated humor or works in a very formal environment they might want to look at different prank options if you this definitely falls into the classic shock value category of humor okay if you work in a more sophisticated work environment or a more sophisticated sense of humor you might want to look for a different podcast everybody i'm sorry but it is a sloppy nominee kevin we have less water on the planet because you clicked that button gab i feel very proud i would put a hundred dollars on the table right now to make that i know you all right everybody we will see you next week thank you so much for watching this is ai for humans sloppies out