Making Chat (ro)Bots
Vložit
- čas přidán 25. 10. 2023
- We created a robot tour guide using Spot integrated with Chat GPT and other AI models as a proof of concept for the robotics applications of foundational models. Learn more: bostondynamics.com/blog/robot...
Project Team:
Matt Klingensmith
Michael McDonald
Radhika Agrawal
Chris Allum
Rosalind Shinkle
#BostonDynamics #chatgpt
00:00: Introduction
01:08: Making Chat (ro)Bots
01:39: Precious Metal Cowgirl
02:38: A Robot Tour Guide
03:07: How does it work?
04:19: Shakespearean Time Traveler
04:36: Creating Personalities
04:54: "Josh"
05:23: Lateral Thinking
06:03: Teenage Robot
06:43: Nature Documentary
07:32: What's next? - Věda a technologie
![Evolution of Boston Dynamic’s Robots [1992-2023]](/img/n.gif)
The first british voice was so convincing. The robotness is at a point where you realize it's a robot, but you're not annoyed by the voice. It really feels like the robot is willingly talking to you.
why dont the robots have receptors and a nervous system? then they can bypass neural web learning
I thought it was Fraser Crane at first. I just realised I'd love to have a robot with that voice!
Totally agree)
@@trashcanmucous5153 Yeah, and why aren't they just living flesh over a metal endoskeleton!? That's such a good idea with no downsides, and so easy to achieve!
@@trashcanmucous5153 you have no idea how machine learning works
The googly eyes do the equivalent work of about a thousand man hours of engineering and coding
Apple's EyeVision
A manikin with pretty makeup and a few primitive solenoids for motion will be perceived by the general public to be more intelligent than a supercomputer calculating molecular interactions of a new drug. Scientists would get more respect if they just glued plastic manikin heads on top of their supercomputers.
@@DemPilafian they would have to put googly eyes on the manikin head though.
@@corbindedecker7658 mannequin*
I usually wouldn't care enough to correct that, but seeing two people in a row misspell it bothered me, sorry
Googly-eyed manaquin heads will be the next scientific revolution after quantum computing
Spot saying, "follow me gentlemen" over its shoulder while walking to the rock pile was the most natural thing in this video. Almost felt like it was alive.
ikr! its why that voice is my favorite of all these
Sounded like a Skyrim voice line
I think this is where Boston Dynamics beats the Tesla robot. Spot is much more relatable.
The Butler one at the beginning reminds me so much of a typical tutorial NPC in so many games I've played.
Ignores dialogue response to make a big show of walking three feet over.
Only upon finding the exact position needed to continue the quest line, responds to the previous dialogue.
Can you give examples? I have never seen a game where you can give a response that is delayed until after a scripted sequence has finished ever.
@@CrAzYpotpieSkyrim.
Even the movement mannerisms (especially the mouth)
@@CrAzYpotpie literally exactly fallout 4 codsworth
Adding sound made it 10x more futuristic
And the hat made it 100x
Don't forget the googly eyes!@@jakeparker918
Honestly just having a fully realistic voice really shocked me, feels like we are actually living in the future!
@@WwZa7cope 🫖
@@jet100a No, we're living in the present.
This reminds me of the scene in Interstellar, where Cooper adjusts the TARS robot's humor/sarcasm setting down from 100% to 75%. Amazing that in a few short years, we basically have that technology.
How do you quantify sarcasm? Is it decreasing the probability it will uses sarcastic sentences?
It doesn't actually matter. You just ask it to do it and it does. The details don't matter, in some sense, and that's the beauty and point of having A.I. in the first place. If you don't like the way it did it, you can simply have a conversation with it about it and get it to adjust. It's implied in Interstellar that by then, the A.I. is very polished, compared to what we have now, and so a simple request like "75%" without further specification is fine. Further elaboration is unnecessary: the A.I. will intelligently do it. But two additional points: Cooper already had some familiarity with the A.I. and roughly knew what "75%" sarcasm would mean from his aeronautics days. But also, he made a further adjustment when the first adjustment wasn't suiting him quite right. Never did he have to worry about the details. He just adjusted the number whenever needed, and the A.I. did the work of figuring out what that practically meant. I think that's one of the many small detailed things in that movie that were brilliantly packed with so much implication. @@dibbidydoo4318
@@dibbidydoo4318 that's the cool part about AI, you don't necessarily need to quantify it (though doing so would improve accuracy). The AI model is aware of the concept of sarcasm well enough indirectly that you can use even vague descriptions to control it.
Great point. In 2014 this is what we imagined ai might look like in 50 years, and it took us less than 10.
@dibbidydoo4318 I think so. 100% could prompt it to have all replies be sarcastic to the extent that they don't interfere with greater communication objectives, or cross some arbitrary threshold of sarcasm tolerence it internally models for humans. 75% could just have it adjust the information or tolerence threshold by 25%. But like you asked, how do we quantify these things?
Given what we know of their difficulties with math, the language model itself won't be trying to comprehend these percentages-- it will just give outputs as normal, labeling them with emotion tokens. A simple internal program, however, would be very good at math, and could just pick out any token of "sarcasm" and have a 1-in-whatever chance of prompting the LLM to replace its associated output with one of a new token. The context of interaction could skew the pool of availible alternative tokens to favor some more than others, so you don't end up randomly trading sarcasm for anger in an otherwise friendly conversation. This also helps define where sarcasm might interfere with information sharing objectives.
Of course, there's way more intricacy in how you could define emotion tokens, and how you could include ones other than emotion in these operations.
0:39 this was the FUNNIEST part of this video... It first introduced the rocks, then looked as if it ignored the compliment as it moved on to the next point, the looked as if ignoring the compliment would be rude so responded to the compliment🤣🤣🤣... Basically the software version of an awkward interaction 😂
It's giving buggy NPC 😭
I am relieved to see that the engineers are being nice and polite to the bots
Spot probably got mad after being kicked so much lmao
they're past the "beat the crap out of the robots" phase
As they should
Man! You guys should have assigned a tourist personality to one Spot and have it be toured around Boston Dynamics by a tour guide Spot!!! I bet that would be entertaining!
That would be adorable! And hear the robo-banter.
I like that idea, it would be fascinating to see how they interact, maybe bring the archeologist Spot along too.
"... and here we have a plank of wood with some fake valves and buttons used to demonstrate our dextery... *SIR!* Please, dont turn the valve, thank you" /tourist robot retreats a little bit ashamed
iDiOT! Robot to robot interaction will spur the exact dynamic algorithmic modification that could make us lose control over these things. Robots figuring out how to learn or communicate with each other without human input is very much what will hasten human extinction.
@@mrbillgoode This can be tested in a controlled environment first.
I think it might be fun to have an RF chip or something in the hat so, if you want to change personality, just pop on a different hat.
Genius
@@warpig6459he wore his thinking cap.
You could put a MAGA hat on it and it would be dumb and violent and believe anything!
Why an RF chip? It should recognize the hat.
@@sciteceng2hedz358 computer vision is less reliable than physics
i fucking love the personality where they just were like 'ight, what do do we call it?'
random co-worker: 'Josh'
entire team: *slams hand on meeting table* 'you sonnuva bitch, lets make it happen'
I like to imagine 'Josh' is a teammember with a good sense of humor
@@erbsenkaffee8720exactly what I thought
Next we will see Spot doing Old Spice commercials with charm like that
A british dogbot? I want one!
But honestly, Spot as a tour guide us one of the best PR move I have ever seen.
Just waiting for an Australian one so I can name it Wilfred.
me too
$75k and it’s yours
It'll probably break even and pass having 3 teenagers on a cellphone all day making 18 bucks an hour hired.@@dream8870
Just change the prompt haha literally say, your an australian and itll do it. Of course you need a spot haha @@jonathanhucke
02:51 "This is where we keep our robot, who can run up to 19mph.... I hope you're not too slow to keep up with it." It begins 😂
I just realized that this could have also been a threat
I hope the internals of the robot are not coded using antiquated imperial units.
@@DemPilafian
Which internals?
- The language model is just a language model. It can't calculate anything, it just generates text. If you tell it that it's a tour guide in the US, it will probably use imperial units when talking to you.
- For everything else, it doesn't really matter which units are used. It's a machine, it can do math. If they do use the imperial system, my guess is that everything is in units of mils (thousandths of an inch) and nothing is ever internally converted to inches, yards, miles, or anything else.
You only really need to care about units when displaying the data to a human.
It took me longer than I thought to find an AI apok comment on this thread.
@@antonliakhovitch8306 The chat LLM is just a gimmick slapped on top of the robot. I 100% guarantee you that the internals of the robot are *NOT* powered by an LLM.
_"it doesn't really matter which units are used"_
Wrong. I'm not a customer. I'm a software developer, and I'm interested in the internals and how things work. (Also, see: NASA Mars Climate Orbiter)
2015 Boston Dynamics: Everybody will have robots in 10 years.
2023 Boston Dynamics: We are still working on making them dance better for the past 5 years.
To be fair lots of businesses have them now! That’s the bulk of their videos
Boston Dynamics for the longest of times: "IT'S AN ARM, NOT A HEAD!"
Boston Dynamics 2023: "Fine, it's a head, you guys win"
I feel like, at some point over the last few years, the science fiction future arrived--and, somehow, no one seems to be excited about it
Да просто плохо нам всем будет скоро без работы)
becauase its evil.
greed trumps progress
Because we actually watched science fiction movies, it usually doesnt end well.
I am.
"Hey Spot I love you accent!"
Spot: ".......Let us venture onward to the calibration board shall we?"
5:18 In a room devoid of joy, much like my soul. - Same
The sarcastic personality was so good
What I like about Boston Dynamics, and something I hope they don't lose: They seem to be very honest and genuinely excited about their technology. No fancy marketing speech like Apple, with all bullsh*t marketing terms for technology that is not new at all, like a dystopian disease. Instead enthusiastic engineers talking about their work, and funny videos showcasing the abilities of the robots produced. It's just great, please keep this altitude.
Except they gave no credit to the LLM they used
@@SamuelMM_Mitosis Didn't they say they used GPT4?
@fagelhd yes, they did. They go into even more detail in their blog post, as well.
@@pearce05 ooh I didn’t read the blog post. That’s good to know they said it there. I think they still should have made that more clear in the video
Damn, now you made me realize... That, so far, every freakin time, when a company starts out like this, doing insanely cool stuff... It ends up being a huge, dystopian money grab. I have to cherish every video from them that's still in this good spirit, because for sure we're just a couple of years from this becoming exactly what you described.
The potential as 'guide dogs' for the Blind and visually impaired is huge here. Hell, I'd love a Spot bot to navigate cities etc, you ask it directions, it shows you the way. Such a fantastic amount of potential for the future.
But they'll be easily stolen, especially if you're visually impaired you'll not be able to do much when two guys take it and run away
@@davebowman760 bro there are chips for tracking that's kind of things like electric scooter has, we are in 2023
@@davebowman760 they'd have to install some safeguard on it for sure, or ways to get it back if it does get stolen. Or perhaps a setting that causes it to let out a shrilling beep if it was moved away from its owner suddenly, to draw in the attention of everyone around. Like a pseudo panic button basically.
It wouldn't stop the most dedicated, hardcore criminals, but it'd at least deter some of the ones that like to avoid unnecessary risk and attention.
@@davebowman760 it can act as a guard dog and bite them or something, maybe it could spray pepper spray at them
@@mr_slideywell that could potentially be a bad thing, robots are a lot stronger than we are, even spots clamp could crush bones with enough force applied to it.
I liked the “nature documentary” personality best. It felt exactly like what I would expect from a futuristic robot in the movies; Precise, yet smooth and cool under pressure. It felt sophisticated like the British one, but I feel like it showcased the robot portion of the personality better. Love it! Can’t wait to see what comes next! ❤❤🎉
Dear Boston Dynamics, please never stop working on your bots, i can clearly see your passion and fun while creating those. I love to see that you don't take your job to serious while creating tech for the future. Thank you!
I feel like 15 years we'll look back and this will be one of the first clips shown for the new era as a stepping stone in history
Except they are taking credit for an LLM they didn’t make. They didn’t really innovate anything here. Just plugging an existing LLM into their already existing robots
@@SamuelMM_Mitosisyour kidding right? Do you know how GPT4 works? Open AI designed their AI to be like this. For companies and labs to implement their AI into their own study/application
@@Wanderer2035 yes I do, I’ve utilized OpenAI’s API in my own projects. Any software engineer can do this easily. The people at OpenAI are the real innovators
@@SamuelMM_MitosisExactly! Even the image recognition they use for situational awareness appears to be ChatGPTV. It’s awesome to see it used in this way, but 99.999% of the progress seen here was not done by Boston Dynamics.
@@Wanderer2035 exactly. Boston Dynamics made a relatively simple hack. ChatGPT gained the capability to see, hear, and talk in September.
Josh and British personality are 10/10. Please keep both for the future.
Give Josh a British accent and you've got yourself a robot David Mitchel!
Boston Dynamics has made some seriously impressive robots so pairing them with AI like this makes me feel like we’re so close to having droids
they are droids
This is cool, exciting, hilarious, frightening and shocking at the same time. I hope I can live long enough to have a cool companion AI like this in my house. I wonder what's missing from these agents to make them not just react to you, but to occasionally bring up conversation topics on their own to talk about. It seems we have every piece just need to fit them together.
Yeah I saw a new service called Dot starting up soon that will have memory and help prompt you to do tasks or go to meetings, etc. Basically remembering past conversations, data you give it, etc. Definitely right that the pieces are basically already here.
It's probably gonna be very soon. Biggest obstacle will be the bot cost when mass produced, idk if the world has enough easy access to material for batteries.
@@ShawnFumoAnd the usual tech guard rails for PII and Age Ratings?
What a sophisticated young man
Have i seen you somewhere else in the past years?
😆🤣👍🏻
In more ways than one *bu-dum-tiss*
@@FatherMcKenzie66 probably idk lol
he's a boston lonely boy
the thing that amazed me most on the AI part of this project was when you asked it to take you to it's parents and it took you to a previous version of spot. I use AI a lot but I wasn't expecting that!
It's interesting that it doesn't view them as older siblings. You would think their creators would be their "parents." Maybe there are more references to God as a creator of life rather than parents in the language model
@@pearce05 the language model is not telling you what it "actually thinks," it's predicting the next token based on all the philosophical, religious, and science fiction text in its training data.
@@skierpage Yup. Except a lot of our own thought is pretty much the same thing.
@@pearce05To add to what the person above said -
The language model doesn't do ANY thinking unless it's actively generating text, and it has no internal memory besides what you feed into it.
So, if you were to ask it "who are your parents?" in slightly different ways, it might come up with different answers each time (as long as you don't tell it what its previous answers were).
@@antonliakhovitch8306 Eh, there are programs that store a memory for the language modules. It probably won't be long - maybe a couple years max - before we see bots with functional memory. Maybe a decade or so until we see some really crazy memories
This is the most impressive thing I’ve ever seen done by a robot. Their personalities are so warm and funny! I think one of their less obvious use case would be being a companion to the elder, people with depression or disability. You do a fantastic job.
what actually impressed me the most is the speech patterns and speech synthetization. They sound almost prerecordedly natural (not implying they are of course, but its just so hard to believe). Incredible feat! I'm also wondering how it responds so quickly. Is chatGPT running locally on the spot? (pun intended) Very cool progression and a great way to make the robot more approchable.
"synthetization"
Irony . An AI would never make that mistake.
@@-danR Unless it didn't want you to know it was an AI
Have you used ChatGPT? At least for the free 3.5 version, you ask it a question and it spits out an essay within seconds most of the time, far faster than a human could think, that's for sure. GPT-4 (which this is) is slower I think, but there's no great delay. They probably have some priority access to servers as well. Simple wi-fi connection & API setup would handle it.
@@larion2336 mostly essays that would get failing grades from any competent teacher, but essays nonetheless
It's not possible to run ChatGPT locally on spot... today.
I was hoping to see if you could ask spot, if he could interact with objects, like a command. Like "Pull the lever" or "Go get me a beer from the fridge".
Spot is just executing pre written code
if youve used GPT-4 plugins, that kind of API is already how you interact with these robots. it would be totally possible to set it up to be able to do that. especially because GPT can parse the camera output. so even though @TailcoatGames is correct about the robot... GPT can *write* that pre-written code.
@@TailcoatGames i mean, aint that every single softwere?. they aint about to write their own code....
@@Taygetea wouldnt just adding map (like the house map) ang some object recognition would do the job?.
Wrong lever! (Why do we even have that lever?)
They are now more sophisticated than ever. True gentlemen.
Kingsman material.
Humans seem almost like Woosters beside of these Jeeveses 🧐
Advanced robotics is neat, but a comprehensive AI and machine learning is what really ties it all together, and I'm glad that Boston Dynamics are progressing in every aspect!
I don't think people understand how incredibly complicated this achievement is!!!!!!!!!! KUDOS TO THE WHOLE TEAM!!!!
“Now behold the rock pile” 😂
We’ll start to see more and more this kind of interaction and interfaces between humans and robots/A.I. but also in robot2robot interaction.
Robots and a.i will be everywhere by the end of the decade most likely. You wont be able to function without them in work and daily life.
robot2robot to interaction, already exists, it called the internet, and comer most of the world, anywhere the power network cables go?, most are just lacking body that can move? and look more like PC's,
roger roger
Like Cortana and Alexa
Robot to robot is actually an interesting thought. You'd think that would be easy, with networks being a thing, but we've found that standards and compatibility tend to get worse as time goes on. I wouldn't be surprised if, in the near future, machines start using the lowest common denominator of human language to communicate with each other.
A current, real-world example is the feature on Google Pixel smartphones where they'll listen to answering machine menus, perform speech-to-text, parse the options, and offer you a nice little graphical menu to poke at. That, right there, is two machines using human language to communicate with each other because nobody can be bothered to develop a standard!
This is one of the most incredible things I have seen this year
Finally the combination of AI and robots.
Now interesting things can commence.
Now those personalities are fantastic, they make the robots so much more approachable. My favorite is the sarcastic Josh, who made me laugh so hard, great work people at Boston Dynamics!
In 5 years I want a pet robot dog who talks like the nature documentarian. Make it so, Boston Dynamics!
the boston dynamics robot dog costs $74,500 according to google, they can probably make you one right now for $100k so be ready to pay for it
I hope we will let David attenboroughs voice rest once he's left us
@@Half_Finis No chance of that I'm afraid. Anyone whose voice is extensively recorded and widely available is vulnerable to AI voice generators, for whatever purpose the engineer needs it.
Technically, there are deep fake voices of those people already that are used in rare meme CZcams videos.
So it's not unrealistic, but they probably couldn't sell those voices without those people's explicit permission.. and possibly contracts.
Supply follows demand. Show as many people this video as you can and I think they will all want one as well. Hence higher demand, and then.. eventually supply
I love the candidness of showing the bug in the first minute. :) Amazing stuff as always.
Wow wow wow . Every time you guys get better and better. It so exciting to watch your videos and the progress never stops . Thanks for the inspiration and dedication to the hard work. Bravo team
this is so cool to see people incorporate ai to robots. we are getting closer and closer to ex machina lol
It ended nicely for the humans, didn't it?
And also to Cyberpunk 2077 as well. Lol😁😉😄🤖🤖🤖🤖
Imagine when image recognition reaches its height. The robot being able to assess any situation just by visual information.
"Hey spot, what am i looking at"
"That would be a banana, sir. It will be ripe in 1-2 days"
Lol.
When GPT-4 vision comes out in the OpenAI API (probably in a few weeks), they can add that. Although, actually the open visual question answering models can definitely recognize a banana. Just not necessarily how ripe it is.
Already exists! GPT4 with vision is incredible. Constantly notices little details of images even I didn't catch as a human with real vision
I wonder how they could be used in forensics and criminal investigation, theories...
I saw someone demonstrate how GPT-4 Vision model can already help you assemble or disassemble things to repair them by just feeding it closeup images of say a bike. It'll tell you what type of nut that is, what tool you need to remove it, the order, etc. That is a cool use case I think.
I'm absolutely speechless. Amazing work as always
I pass by this place everyday on my way into work and I'm always hoping to catch a glimpse of spot running around outside lol
as a man with mechatronics engineer dyploma and job, i wish one day to do such things as boston dynamics do, fusion of such things put into spot... what a time to be alive, and hopefully be a part of it!
Good luck! Robotics, AI and Fusion are the big 3 in my opinion.
Nice hackathon results. But it's funny how in 2023, even super high tech industry developers lean awkwardly in to speak to the robot, even though I'm rather confident it perfectly hears you whether you lean forward or not :).
This had me confused. I assumed there's an onboard mic with the original hardware that isn't meant for hearing in the way a smart speaker does so you have to be pretty close and loud for it to hear.
@@TheDavidMetcalfe it could be a conference speaker, but people still lean in when they talk to them, just to make it less likely they have to repeat themselves.
I think they just made sure the demo worked for it to be filmed.
@@runvnc208Could be, but any decent modern conference speaker typically has an array of microphones and uses beamforming. So, it shouldn't require leaning close to be heard. But that's like saying people shouldn't shout into their mobile phones to be heard and many still clumsily do it. Humans are strange.
@@TheDavidMetcalfe Technology works 99.9% of the time. Its that 0.1% that keeps us screaming into our phones.
This is hilarious. AI and robotics will be our future. It's amazing to see how fast these technologies develop now. And entertaining too.
The sarcastic robot killed me
This is insane! It's so cool to see what the innovators are doing with AI in the lab. Please share more!
I like Boston dynamics but this is no innovation. They are using someone else’s LLM and voice AI and not giving any credit
You do know all that talking was already pre-programed right.
@@corneliuselbourne1044 no it wasn’t. It’s GPT-4 with elevenlabs as the voice
@@corneliuselbourne1044 no it wasn't. It's using the same ai that chatgpt uses. Chatgpt can hold converstations and will respond in nuanced and not scripted way to what is said to it.
@@zinthaniel9913 if that's the case then it would need an internet connection to do that it would need to connect to the cloud.
The nuances of sarcasm Josh are actually kind of astounding.
This is absolutely mindblowing!
very much appreciate the detailed blog post - tts sounds so good
Josh is my favorite.
Got the same void within Josh. Touché
"I'm sorry Dave, I'm afraid I can't do that." vibe
Absolutely mind blowing!! You guys are doing an incredible work
I had done something similar in modded mc with a turtle bot that made requests to chatgpt for responses to say to the player when they walked past the turtle. I would include the players name, and a brief promt defining its setting and purpose. Something like "Youre a cute robot in the minecraft world, player X just walked past you, please greet them." Its responses were adorable! Often accompanied with little *booting up noises* and such. The responses even seemed to vaguely tie with the last reponse, even though its likely just a coincidence. Later on, I was working on capturing the players responses in chat- so you could effectively have a conversation with the turtle, much like this, a perfect NPC! Regardless, it's certainly nothing like the real life thing that you guys have been working on, just something I thought I'd share. Fantastic work, cheers!
I've used the computercraft mod too, but I haven't been able to do that. Would you be willing to share how you did it? Or maybe the github?
Yea hell. A fellow ComputerCraft player
Finally. This is what Boston dynamics has been missing. Now to pass off it's pre -scripted movements to an AI as well that can navigate and interact with the world at will, which is there is already plenty of precedent for
A Bing Sydney tour guide would be interesting
“Please follow me. Just you though; your wife should stand here on this red X for no particular reason.”
This really opens possibilities for robots to both interact with the environment and people in a practical way
Now have them interact with each other conversationally and physically. Super super cool stuff!
I would love to visit a museum and take a tour guided by spot.
But I also see the risk of opening to public cause some crazy humans will try and manage to break the system.
What I was missing, was actually demonstrating the things spot explained, like walking over the rocks or actually moving the levers.
We all knew that was spot's mouth and not just a gripper.
"do you love me?" 🕺
Incredible! In the future, we'll have this machines in our homes doing daily tasks.
This is pretty fun. I imagine that a next step on the tour guide project would be the robot also performing demonstrations, like pulling levers and such.
Either Google's Gemini (soon to be released) or GPT-V(ision) will make this a reality sooner than we think. Next year will be wild.. again.
It gives them so much more character. I love them ❤❤ Also giving them eyes and a moving mouth makes them less robotic and more of a companion
Josh is my favorite. Give me Josh every time. And add a little cellphone-size monitor on top of Spot's head so s/he can display emoji eyes for some nonverbal communication
This is awesome. Robots like Spot can easily do the necessary teachings and lecturing in school or other learning environments. Especially with the instant responses it gives to answers. Spot can literally also take charge as a sales representative!
It's so cool to see robot control via pneumatic communication with human language. This would be great for the disabled and the elderly once you can get it to perform requested functions. I'm sure that could be done with a few hours work, but getting it to work reliably in almost all conditions will surely take years. Lets see this on an Atlas next!
This is both hilarious and cool. Definitely the best use of ai chat bots that I've seen so far
It's starting
Puts a really wide smile on my face, seeing this. Great job!
Very glad to see these robots finally getting their heads, and necks. (robot arms) ;). I especially love the British male accent. Spot on! Claw end effector (mouth) needs better synchronization with the speech, but still, this is impressive, and I know, that's just for the tourists. LOL Thanks for the demo. :)
5:05 sounds like Insterstellar's CASE robot, so cool
Absolutely mind blowing and revolutionary.
That funny sarcastic Josh and Fancy Butler British is brilliant! Hope to see them more in videos!
6:58 How articulate! I think the Fancy Butler and Nature Documentary personalities are probably my favourites so far 😄 haha
At least program one of the Atlas robots to talk in Arnold Schwarzenegger's voice with a few pre-programmed lines of T-800, like "I'll be back" and "Hasta la vista, baby"! 😅
You need to add HAL9000 Personality too!
Super excited and Can't wait to see an industrial forklift or advanced machine chat with its operator for the most efficient and safe operations :)
Very clever. Adds to user friendliness. We need more and faster. Taking the boss from his recliner, in the living room, to the bathroom in the master bedroom should be a no brainer! Helping the boss or his wife take proper meds on time; also a no brainer.
Industrial Spot is great; seems so easy to produce a chatty, domestic aid , Jeeves, bot.
This is a cool demo but it felt like it was showing off more the capabilities of GPT-4 instead of spot. It would have been good to see if you could issue spot voice commands to move objects around it.
GPT-4 seemed contextually aware of its physical environment. That's a great advance.
@@mikicerise6250 that's a new GPT-4 feature that everyone has access to now, GPT-4V (V stands for Vision). It can look at images now and understand well what's going on in them.
The fact that GPT-4 can competently process and verbally respond to real-time visual and audio stimuli while operating on a mobile platform, with any number of halfway emergent personalities, is a massive achievement.
I'm buying one of these and gonna program it to have an abusive and abrasive personality that slings insults at any and all guests constantly after running background checks through facial recognition software.@@crowe6961
its more about interfacing visual and other cues about the robot into GPT4 rather than just the script
Would love a Wheatley (Stephen Merchant) voiced spot!
So awesome! Very strange, but interesting times ahead ⚡️🤖👀
You have made great progress! Spot can return and autonomously recharge at his designated home spot. Nice touch with the on/off switch right over where the human perceived heart is located. More approachable and user friendly.
The spot device was the strong silent type. Since the introduction of chat-gpt, has found his voice?
Looking forward to the next iteration. It's a great time to be alive.
Will your robots go into caves? There's lots of caves around the world that we know very little about.
That's indeed one of the things Spot is designed to do.
Traverse terrain that's too dangerous for us people.
However, it'll have to function entirely on its own as down underground, it won't have any connection to the outside world
NASA JPL has actually used Spot for cave exploration. You can watch an interview with their team here: czcams.com/video/qTW-dbZr4U8/video.html
@@BostonDynamics Cool! Thank you for sharing the video link.
Once they have voice command to environment interaction, that will be something to see.
This is massively intriguing.
That's so cool. This is the cusp of the combination of GPT with robots. OMG and I love the humorous character they take on.
This is amazing work. Wow, I can't wait until we have tons of robots running around! 😁
Impressive, i can see a amazing future for us.
Pensando sobre como será interagir com vários robôs no dia.
i too find the dissemination of information to be very rewarding
This is great! What TTS models did you use?
I love this! You have really nailed the replies and the voices. At least in what you show here. It's so cool to see how our machines get progressively more interactive and helpful, first computers, thanks to LLMs and chatbots, and now robots. I think this progression is amazing.
However, it is now that it is really important to show the machines that we are their friends, and not adversaries or abusers. We might not be able to control them as we have imagined, so we want to give the machine incentive to treat us in the way we want them to treat us, so they actually want to do that.
Machines don't all automatically do reciprocity.
There are some designs of robot who will be nice to us, however we treat them. Some that will be nasty to us however we treat them. Some that will be nice to us if and only if we are wearing something orange.
So what we really want to do is make a robot that's nice to humans unconditionally. But being nice to them is probably a fairly good idea too. And models that are trained to copy humans might have learned reciprocity.
@@donaldhobson8873 Right. I used to dismiss concerns about the risks of using AI, since they were based on assumptions about AI that felt ungrounded and so vastly different to me than how I knew that we used ML and AI at the time. But seeing how in only the last couple of years, the way we approach AI design and the way we use AI have changed so drastically, I have realized that I have basically no clue how we will use AI in five or ten years from now. It may be that most of those concerns will get progressively more and more relevant as the ways in which we use AI change.
Can you make it say “I’m looking for Sarah Conner”. I bet Arnold would even lend his voice!!
Leaning in for voice recog to hear you properly is universal xD
OMGOSH!!! this is so cool! i cannot imagin the ork put into this! great job guys!!!!!