You imagine you crack fusion with it or you know room temperature superconductors and you know batteries that are optimal that opens up you know suddenly energy becomes free or cheap then that has huge consequences on you know resources like freeing up like you could do more space travel M asteroids
Maybe becomes feasible all of these things right and then suddenly the nature of of money even changes right so I’m not sure people are really understanding like I don’t know if company constructs would even be the right thing to think about at that point
So the open AI backed robot is ging some traction CEO and founder burnt bornich says new progress update on the droids dropping in 4 weeks looks like morx Paradox might be false and we just didn’t have the data what does that mean we’ll dive into that in just a second
Also Ted Chow who’s working on robotics for Google deepmind says there will be three to four massive news coming out in the next weeks that will rock the Robotics and AI space adjust your timelines it’ll be a crazy 2024 I have a feeling it will shock the entire
Industry in fact just a hunch Dr Jim fan drops a career update is founding a new group called gear which stands for journalist embodied agent research 20124 is the year of Robotics the year of gaming Ai and the year of simulation he finally also got recognized for
Something that for some reason I feel like a lot of people have been missing I did multiple videos on it it’s just not getting the traction that it should be but people are finally waking up we’ll look at that in just a second also opening eye drops some massive news
About the GPT store but first I have to touch on the whole Google Fiasco I didn’t put out a video on the DAT that it was happening but Google has been changing some prompts to be different than what users want them to be I mean
Here are the images of the pope of the founding fathers image of a Viking image of an American woman image of a British woman image of a medieval Knight an Australian woman and here’s their rendition of Chess that last one was a joke obviously but here’s Demi saabi so
He’s the founder of Deep Mind which is part of the alphabet company so the same kind of umbrella that Google under not directly part of Google but looks like they’re merging more and more he just went on an interview and explained kind of what was happening there behind the
Scenes with these image Generations so we’ll take a look at that as well so the this person that made all their tweets private I believe this was this was Jack Kik who is a Google employee who is the senior director of product for Gemini experiences so he’s likely behind a lot
Of the decisions that kind of went into this he was responding to the backlash by saying well no it’s not true that’s not happening here’s the issue with that this person I think nailed it perfectly they’re saying the actual problem is the lack of transparency of what you do with
The prompts and how you’re manipulating the output just be transparent and none of the backlash will occur I understand that the intention is good but if the ethnically ambiguous gets added to a prompt please let the users know that instead of gaslighting them David Shapiro posted this and I think it
Really also cuts to the heart of the matter and here’s what he’s saying one user consent matters hijacking your intentions as a user of any technology is wrong enforcing a particular worldview on users is flatly unethical yes racism is bad but cramming one for-profits company’s view of morality
Down the user throats is dystopian AF and I think that dystopian AF is an excellent way of putting it corporations setting themselves up as Arbiters of morality is Flatout cyberpunk morality and ethics is up to the people not the corporations and not the government looks like we have overwhelmingly people
Agree looks like almost 90% agree strongly or mildly there’s been some people noticing online this week that Gemini doesn’t seem to create a white man if you ask it to or if you try to depict uh figures from history it um it sort of doesn’t want want to do that or
We’ll sort of do it in a historically inaccurate way you know I think with this this is a good example of these nuances right the historical accuracy absolutely we want that um so we need to fix that versus when you have a generic prompt obviously then there things are
Universal so for example if you said you know create a picture of a person walking a dog you’d want a universal you know depiction but then others that you know historical events or historical figures then perhaps that should be narrower and and as I said we’re continually improving our models based
On feedback now here’s an interesting take so apparently this person Yan was once the CEO of Reddit so he’s saying Google’s Gemini issue is not really about woke diie and everyone who is obsessing over it has failed to notice the much much bigger problem than it represents to recap Google injected
Special instructions into Gemini so that when it was asked to draw pictures it would draw people with diverse or non-white racial backgrounds this resulted in lots of weird results where people would ask it to draw pictures of people who were historically white like Vikings or 1940s Germans and it would
Output black people or Asians Google originally did this because they didn’t want pictures of people doing Universal activity like walking a dog to always be white reflecting whatever bias existed in the training set so this person continues this is not an unreasonable thing to do given that they have a
Global audience maybe you don’t agree with it but it’s not unreasonable Google most likely did not anticipate or intend the historical figures who should be reasonably be white result now he continues and he talks about the three LS of Robotics by Asimov The zero flaw and some bad things that happen in those
Books brings up Elija ovsky we’re not going to go down this path quite yet but his point is that AI can have unpredictable outcomes what are the the cases it brings me back to something like this the actual problem is the lack of transparency of what you do with the
Prompts and how you’re manipulating the output just be transparent if the AI is outputting something that doesn’t align what you think it should show if you just start manipulating the user’s request without letting them know that you’re doing it and not being open about it whatever Noble purpose you thought
You had becomes dystopian AF just don’t lie don’t be evil as I was finishing up recording this video this popped up in my feed so this is Google’s blog saying Gemini image generation got it wrong will do better they’re saying that it’s clear that this feature missed the Mark
It’s inaccurate or even offensive basically having a diverse group of people if you’re asking for somebody walking the dog however if you’re asking Gemini for a specific type of person such as a black teacher in a classroom or a white veterinarian with a dog or people in particular cultural or
Historical context you should absolutely get a response that accurately reflects what you ask for and so they’re saying so what went wrong in short two things first our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range and second over time
The model became way more cautious than we intended and refused to answer certain prompts entirely wrongly interpreting some very anod prompts as sensitive so some very inoffensive prompts and sensitive these two things led the model to overcompensate in some cases and be overly conservative in others leading to images that were
Embarrassing and wrong they’re saying we did not want Gemini to refuse to create images of any particular group and also Elon Musk posted this a senior exec Google called and spoke to me for an hour last night he assured me that they are taking immediate action to fix the
Racial and gender bias in Gemini time will tell and I’m kind of with Elon this one time will tell I want to believe that this was a innocent mistake caused by over tiing the commands on the AI model but I think all in all this was a
Good thing because it showed to a to a lot of people why it’s important to have transparency with AI why it’s important to understand who is training these models what their viewpoints are maybe this was a simple mistake but if in the future it gets used for malicious intent
How do we know do we have visibility into that decision process so I agree with musk here time will tell often times the argument against have open source AI is the chance that it will be taken over by Bad actors what protections do we have against the big
Close Source centralized AI from being taken over by Bad actors in other news so this is Logan GPT the Ambassador for openi and AGI he’s saying great news for GPT Builders and users we just launched a big GPT store update with GPT reviews Builder social profiles a new about page
With ratings categories number of conversations etc etc so now you’re able to rate gpts provide private feedback directly to the Builder and an about section in the Chad gbt store here’s Nick Doo so he created the grimoire the number one coding wizard GPT so looks like it’s got over 200 reviews so this
Is it right there in the GPT store looks like it’s saying 1,000 plus reviews now now I’m curious are you using gpts are you building building something on there when it was first announced and I was looking at it I thought this was going to be the next big thing the next big
Wave however I’m not quite sure how much of an impact it has made now it seemed like since then we had a lot more news seem to indicate that everything’s moving much faster now like this might have been a much bigger deal if AI progress moved slower but with the way
Things are going this was just maybe this was just a much smaller thing that we first anticipated or is it heating up now because there’s more ratings and potentially people starting to be paid for how many users use their gpts I’m not sure let me know in the comments
What you think are you excited about gpts still did you completely forget that they exist I’d be curious to know now the other big news here is robotics so Dr Jim fan is announcing that he’s launching gear generalist embodied agent research so this is kind of under the
Nvidia umbrella we believe in a future where every machine that moves will be autonomous and robots and simulated agents will be as ubiquitous as iPhones we are building the Foundation agent a generally capable AI that learns to act skillfully in many worlds virtual and real 2024 is the year of Robotics the
Year of gaming Ai and the year of simulation we are setting out on a moonlanding mission and getting there will spin off mountains of learnings and breakthroughs so this is the page on Nvidia so this is the other person that’s going to be on the team one of
The main people on there it seems like I believe he was working with Dr Jim fan on Voyager the Minecraft AI as well as on on the Eureka paper which we’ll touch on in just a minute here and so this Nvidia gear team will be led by the two
Of them and the mission is to build these Foundation models multimodal Foundation models LMS for planning and reasoning Vision language models and World models trained on internet scale data sources we also have general purpose robots robotic models and systems that enable robust Locomotion and dextrous manipulation in complex environments Foundation agents in
Virtual Worlds these large action models that autonomously explore and continuously bootstrap their capabilities across different games and simulations simulations and synthetic data and this is the recent thing that we’ve been talking about with things like Sora emerging simulation infrastructure and synthetic data pipelines for large scale learning and
Their hiring so I got to say this is this would be a very exciting opportunity in part because as Dr Jim fan says Nvidia has enough Capital to solve robotic Foundation model gaming Foundation model and Genera simulation all at once he’s saying our new group is perhaps the most well-funded embodied AI
Lab on Earth so I would not be surprised if Nvidia indeed is the company that’s going to produce a lot of these breakthroughs and also Eureka was named the top 10 Nvidia research projects of 2023 we’ve covered it on this channel it was quite exciting they taught a robot
In a simulation to twirl a pen in its fingers similar to how you used to have kids in school kind of do that if you recall which was just an insanely hard problem to solve previously but it gets even better than that in order to train these robots in simulation there’s
Something that’s called a reward function so basically if you’re training a robot to you know walk have some sort of a walking animation to get from one side to the other you might give it rewards for for example you know moving forward some distance you might give
Some penalties for falling on its face whatever the case is you write out an equation that gets it a adab boy every time it does something right and a you know don’t do that again if it does something wrong kind of similar to some of the things we use to perhaps train
Animals reinforcement learning you know give them a little treat when the dog sits down on command something like that now we humans are pretty good at writing those reward functions up to a point we’re good at writing them when the thing that we’re trying to get the robot
To do is pretty simple like balance something or run across a screen or open a door or drawer or something like that but were’re not really good when it comes to doing something that is more complex like spinning a pen in its fingers so Eureka tried something new
And said you know we humans were not that good at writing this code so you Chad GPT GPT 4 you do it and we told GPT 4 what to do the task description that we’re trying to achieve to make the shadow hand spin the pen to a Target
Orientation then we gave it the environment code sort of like the code behind this simulation and so GP bt4 wrote out various samples different sort of versions different approaches to doing this and that was fed into the Isaac JY the simulation the results of that was taken out of the simulation and
Given again to that same GPT 4 with feedback and said here’s what you told us to do here’s how well that worked and based on that it tried to rewrite it tried to improve it sort of like iterate upon it and this Loop continued and surprisingly Eureka outperforms humans
Across all tasks in particular and this is the really interesting part Eureka realizes much greater gains on high dimensional dexterity environments meaning as humans get worse at doing the very complex tasks at writing these reward functions for very complex tasks Eureka does well on those tasks in other
Words it’s better at teaching robots how to do complex tasks than humans can so by the way this Isaac Jim so it’s a GPU accelerated f physics simulator that speeds up Reality by a,x so gpus are the Nvidia chips that they’re so famous for and so just imagine a universe of these
Like little robots that are floating around trying to land right stretching out for infinity or trying to catch this egg on a plate and this simulates kind of our physics all the friction and gravity and everything else momentum to kind of simulate our reality but it’s
Able to run it a thousand times faster and so Eureka you know in that environment by writing rewards for that environment Eureka is a superhuman reward engineer the agent outperforms expert human Engineers on 83% of The Benchmark tasks with an average Improvement of 52% and it’s much greater gains on tasks that require
Sophisticated high-dimensional motor control and and this is interesting this is kind of a big deal I think so Dr Jim fan here is saying to your surprise the harder the task is the less correlated are Eureka reward rewards to human rewards in a few cases UA rewards are even negatively correlated with human
Ones while delivering significantly better controllers what that means is as the tasks get more and more difficult not only is Eureka better than humans it also comes up with novel Innovative completely creative and different ways of solving those tasks or writing those reward functions than humans do so when
People say oh Chad GPT is just spitting out what it’s learned it’s not making anything new how do you explain this it’s better than humans in ways that humans were not even like creative enough to to come up with those Solutions they’re new they’re novel and they’re better and here he Compares it
To Alpha go where it made brilliant moves that no human go player would make AI will discover very effective strategies that look alien to us so this to me was huge because this really reinforces this idea of AI something like GPT 4 being you know out of the box
Better than us than humans at training these robots in a simulation which then these simulations from what we’ve seen they translate very well to the real world so it’s not like this hand that’s training in a simulation is now only able to do that in the simulation you
Take out that data you put it into a robotic hand with the same characteristics and it’s going to be able to do that so AI will be training the next generation of AI and robots so this is Ted Chow so he’s saying how there will be three to four massive news
Coming out in next next few weeks that will rock the Robotics and AI space he says he believes that all the pieces we need for a modern attempt at embodied intelligence and keep in mind so he’s at Google and deep mine here he’s mentioning how much better the data
Collection for robots will be he’s an open source system it’s 400 bucks it’s from Stanford that’s design to democratize robot data collection how do you collect that data well here’s a human kind of just doing those motions there are two cameras on there recording an mp4 interestingly it looks like they
Ed two mirrors there to kind of create a stereo effect to kind of have just more angles on what they’re doing without necessarily using more cameras and here’s the opening ey back robot saying that there’s going to be a new progress update on the droids in 4 weeks and
Looks like morx Paradox might be false and we just didn’t have the data so Mor’s Paradox is the idea that contrary to traditional assumptions requires very little computation but sensory motor and perception skills require enormous computational resources meaning it is comparatively easy to make computers exhibit adult level performance on
Intelligence tests or playing checkers and difficult or impossible to give them the skills of a one-year-old when it comes to perception and Mobility so he’s saying looks like that might be false and we just didn’t have the data so can’t wait to see what these guys drop
What the new breakthrough is going to be before you go here’s a couple clips from the Miss saab’s interview where he touches on some pretty interesting things take a look do you think the world is ready for something like AGI to show up it’s something that’s going to
Affect everyone in society I think there are questions on International cooperation I would like to see a lot more of that unfortunately the geopolitical nature of the world right now is not very conducive to that so that’s unfortunate timing and then we need to of course accelerate our
Research into safety guard rails control mechanisms and also philosophy too like what do we want from our systems kind of and and ethics right and philosophy of like it’s kind of deep philosophical issues like what do we want our systems to do how what values should they have
And I think Society will have to adapt about what it is you know we want to do in a society where we have ai systems that are able to do very useful things for us maybe we have abundance because of that because we cracked things like energy problems things like physics and
Material design so there should be a huge plethora of amazing um benefits that we just have to make sure are kind of equally distributed you know so everyone in Society gets the benefit of that and then you know I think incredible things might be possible that sort of written in science fiction books
Books like the culture series by in Banks and so on it’s always been my favorite since
