I am very impressed by the clarity of writing in this video. If I do not sufficiently pat you on the back, you may join in without any fear of being accused of having a big ego. Well done!
This is one of the best explanations of Artificial Intelligence that I have ever seen. In the human brain, neuron paths will change according to learning and experience. Under the right conditions, a neuron will “fire”, but only to a specific other neuron. Weights and biases in computer neural networks act in a similar manner. At least that is what I understood from this video. Very well done.
@@TheConedera Here is an even deeper explanation into what you were saying: AI, or artificial intelligence, is a branch of computer science and engineering that aims to create intelligent machines that can perform tasks that would typically require human intelligence, such as visual perception, speech recognition, decision-making, and language translation. One popular approach to creating AI is through the use of neural networks, which are modeled after the structure and function of the human brain. A neural network is made up of layers of interconnected "neurons," which are simple mathematical functions that are activated or "fired" when they receive input. Each neuron receives input from other neurons through pathways called "synapses," which transmit information between neurons. The strength of the connection between neurons is represented by a value called a "weight," which determines the amount of influence one neuron has over another. Additionally, each neuron has a "bias" value, which is added to the input before it is processed by the neuron's mathematical function. The bias term helps to shift the activation function to the left or right. The training process of a neural network starts by providing it with a set of input and output pairs, called the training dataset. The input is passed through the input layer, which then passes through the different layers of neurons in the network. As the input data passes through the network, it is modified by the weights and biases of the neurons, until it reaches the output layer. The output of the network is then compared to the desired output, and the error is calculated as the difference between them. This error is then propagated back through the network using an algorithm called backpropagation. The backpropagation algorithm calculates the gradient of the error with respect to the weights and biases of the neurons, which is used to update the weights and biases in the opposite direction of the gradient, to reduce the error. This process is typically done many times with many different examples in the training dataset, and the weights and biases are updated after each example, until the network reaches a satisfactory level of performance. The goal of training a neural network is to find the optimal set of weights and biases that will allow the network to accurately predict the desired output based on the input. Once the network has been trained, it can then be used to make predictions on new, unseen data. The prediction process is similar to the training process, but the weights and biases are fixed and don't change. The input data is passed through the network and the output is the prediction of the network. It's worth noting that the architecture of the neural network, the number of layers, the number of neurons in each layer, and the activation function used, also play a crucial role in determining the network's performance. The training dataset also plays an important role, it should be large enough, diverse and representative of the problem to be solved.
Born in '43, educated in EE, worked with computer hardware and software until '02, completely incapable of comprehending most of the technical details of AI. That's me. Go figure.
You think most of the people in the machine learning field feel that way because not one person makes the whole? Seems like ai is a collective technology where lots of specialties come together. You may build the circuit for real world data collection however someone else takes that data and trains a neural net. I am a hardware person and I am barely scratching the surface to python and data set training. I believe if you dove into the technology you would grab the concept well enough to build your own functional ai creation.
@@idus Perhaps I should complaining about not understanding Crypto Currency, and that I don't understand why Block Chain so-called technology is so all-encompassing and safe. The real and core problem is that I'm too far gone (old) to communicate with people who call table links block chains. Anyway, it's not my problem. I'm kind of happy despite my ignorance. Oh, and training a neural net is is sort of comprehensible to me probably because it had no precursor technology with a different set of new vocabulary to needlessly confuse me. It seems simple and I think I may actually understand it. But heck, thanks for taking the time to read my comment.
There was a massive breakthrough in the 2016-2018 timeframe. You can thing of this as trying to build a table surface when you haven't built the legs yet and simultaneously trying to build the legs while the table doesn't exist yet. The breakthrough involved something of a Monte-Carlo approach allowing the differing levels of understanding to build upon each other.
@@jefflittle8913 - Perhaps starting at both ends and working back towards the middle is good but for me, the two ends are never connected to the same middle. Hopeless..too old. I used to be familiar with the term "Monte-Carlo". Now it's just a 4 syllable noise. And I can't seem to treat probabilities mathematically any more. Shucks
1972 to 2011 was AI winter. I am a hands-on AI enthusiast. Understand the need for flops for computing with large image sets. DOJO architecture and underlying things are the future. We always looked at Compute as limited. Compute is measured by FLOPS. Tesla's AI team broke the barrier by introducing Configurable 8, which is compatible with IEEE 754. The way the memory was addressed changed in DOJO, giving unlimited computing. I see this will be a big game changer; till the world comes out of its slumber, they will scratch heads, and computer vision helps human vision. Hats off to Elon and Tesla masterminds.
Having FSD beta in my Model X, I think autonomous driving is at least 3 years away. Weather is the biggest problem along with pop up tasks like driving through a lane closure where each direction alternates from a worker directing traffic.
..I really want FSD to work as replacing every ICE vehicle with an EV is not the answer and FSD will increase vehicle utilization so everyone does not need their own vehicle. The problem I see is people and governments excepting FSD will NEVER be 100% and the vehicles will still get into accidents, especially as you mix FSD vehicles with human-driven ones - the problem is we accept people-killing people on the roads but will we accept an FSD driven vehicle killing people?
Congratulations on your HONESTY about Telsa FSD vehicles being at least 3 years away. I think Elon and Tesla should go to jail with the hype they make about FSD vehicles, it gets many Telsa drivers getting sloopy with their driving by giving it over too much to the FSD mode. I just heard last week that The US government is preparing to take action against Telsa for its FSD mode and claims. There are RU-vid videos with the FSD Telsa drivers sleeping at the wheel. They should get a dangerous driving charge by the police.
@@UncompressedWAVmusic I really like how he charges people $15,000 for beta testing his FSD feature... after promising everyone they could make $30,000 a year with remote taxi service... LOL anyone who bought this has a class action lawsuit for lost wages every year they had the thing
There's really no telling when full fsd will actually be released. I'd have told everyone when I got my Model 3 in 2019 it would be 2-3 years away at that time...but its still not close and makes lots of mistakes. It's a very polarizing topic because lots of people have paid really good money, and to date, only received a product that you have to monitor very closely so it doesn't make mistakes. But what I've realized during this time, is that it is improving. I've all but given up hope that "it's right around the corner", but I do remain hopeful that if anyone can do it, the tesla team can. As to when this might be, it could be a few months away if theres a huge breakthrough...but it could be 5+ years away if it continues the past 3 years of slow advancement.
I feel FSD is reachable within 5 years, especially with DOJO. Waymo is already having robo taxis in some cities like SF and their FSD tech is light years away from Tesla. The main thing that's stopping Tesla from releasing robo taxis just like what waymo is doing is all the red tape and lobbying that's needed.
„Great video on AI! The explanations were clear and concise, and the examples really helped to illustrate the concepts. Keep up the good work!“ - ChatGPT
Spoken like a true warrior Skirgailia aka Elon Musk aka Voldemir Zelensky aka Benjamin Netanyahu aka Dark Prince Harry aka John Hagee aka Benjamin Franklin aka George Washington aka Martin Luther aka anti-Christ John xiii aka Skirgailia. And you didn’t tell your AI students about the meaning of the scissors, which were placed by Me the day before this what some may call a miracle, because you don’t want to tell your students that yes, there was a miracle. For those who See what I SEE And so it is.
As it has been said, it's hard to build and bring to market. Never been done before, but with this much effort, it will happen. Imagine in the 1950's going to a world's fair and seeing a box that you put your food and within minutes a bell rings and a hot dinner comes out. It had never been done before and now almost all of us have a magic box that can make popcorn in less than 3 minutes. A.I. self driving will be just like that, we will never need to get drivers licenses. A small club of people will have beautiful cars and may be the only people left to own and drive their own cars but maybe on closed roads for safety. ;)
@@eriktompkins7137 in my opinion More likely that human driving will be done by crabby old men only on weekends, in ICE CARS, at designated courses, including racing. And the bystanders will politely applauded their efforts.
@@eriktompkins7137 in my opinion More likely that human driving will be done by crabby old men only on weekends, in ICE CARS, at designated courses, including racing. And the bystanders will politely applauded their efforts.
@@eriktompkins7137 in my opinion More likely that human driving will be done by crabby old men only on weekends, in ICE CARS, at designated courses, including racing. And the bystanders will politely applauded their efforts.
Imagine the value of the data alone that Tesla is gathering…let alone it’s computers…I’m guessing the Dojo would be useful for all of Elon’s companies, among many, many others. Brilliant ❤
The way Elon multiplies and compounds the achievements of each of his companies is right in line with the Dojo computer. Can you imagine the uses SpaceX will be able to tackle with its ability.
@@carholic-sz3qv Its useful for complex rocket equations and simulations in different environments/worlds ,the Tesla robot, processing info from studying billions of neurons with neuralink etc
Just want to add something here! :) I studied AI in Uni and it might be a bit misleading how you present the information here. The Neural network tesla uses focusses mainly on interpreting the environment as accurately as possible. The actual steering of the vehicle is comparable with collision avoidance in a computer game. So random jerks on the wheel is just a result of incorrect environment readout. The actual neural net is not directly connected to steering, as in there is no drive straight neuron. Other than that! Love the simple explanation. Edit: Elon Said in the latest AI Day In the Q&A that its moving in this direction though.
Interesting idea except that Twitter is still slanted left and more negative than positive responses…an ai would learn to be a narcissist on Twitter and not a good human!
We need more computing power but we also need to vastly improve our algorithms. The human brain does not use back propagation. It uses something more efficient and requiring less computing power for better results. A three year old child who has never seen a cat or a dog before will need only a dozen or so examples of each in order to distinguish between the two. An AI, using current algorithms will need thousands of examples of each to do the same job.
Actually this is not an error as the computer is doing the best it can do with the input given…. Technically, error is the wrong word, I would rather say that it is an unproperly trained model AI… or something along this lines
Gives me ideas for some interesting combinations of technology,ai, and organic possibilities. And a couple of others areas of science with assistance from ai there will be some remarkable leaps and bounds in our future.
1. To the Quik: It's Apparent "approximation to errorless" is an [Approach] that should be [Reconsidered]! 2. What [Approach] should be [Used]! To "Easily Locate" an [Object & its Coordinates] "under" an [Irregular Curve]! [To Gather "Accurate" Data]! 3. To Note: 100% Accuracy! i.e. The [Surface Area of the Earth], "Streets & Highways Included", has [Curvarture]! Stay Safe! Tesla #1! 4. Hedwegg where [Speed & Determination] make the Difference.
Vague on the technicalities, or straight away wrong. Tesla doesn't actually retrain the entire Dojo, but only part a time, layer by layer, as needed. This divide&conquer is much more efficient.
Elon's Giga Stack: dojo, starlink, neuralink, twitter, blockchain, autobidder, megapack, solar, tesla bot, robotaxi, robo-semi, robo-ship, and later Muskville on Mars. Let that sink in!
This super computer has the potential to significantly impact the field of image generators by providing powerful computational resources that can accelerate the training and generation of high-quality images. Imagine discord AI bots generating images in a split second.
Mining the miners. The gold rush gold was good but the pick axes and shovels made wealthy merchants. Cities, states and highways, trains planes and ships. Mine the miners, Elon!
3-4 years until fsd. Dojo will accelerate the machine learning feedback loop and give Tesla a significant and perhap insurmountable lead on any competitors unless there are other AI model breakthroughs that reduce the training compute load. The other question is this. Is the current neural net hardware inside actual Tesla cars sufficient and robust enough to perform the FSD task, or will it need to be upgraded to something more capable. The really big step will be when Tesla can upload the AI smarts into its robots. It's pretty amazing that something humans find so simple is so incredibly difficult for AI to master, but eventually FSD will happen. Perhaps any AI model will need to learn from the daily human perspective before it can master FSD.
ehm, a human has been in development for... MILLIONS of years? ;) A.I. is how many? (not to sound arrogant, I know what you mean and how it looks like but hey A.I. has only just peek-a-booed.
@@particleconfig.8935 And Elon Musk has been lying about FSD ever since it first came out. The US government is preparing to come against Telsa for its FSD scandal.
@@UncompressedWAVmusic scandal? Tell me.Well, yeah it's sort of a Biden thing to not acknowledge the actual state of things and attack Tesla with not classifying their cars for tax redemption. It's kinda a thingy isn't it. But FSD will come out (didn't you see it drive perhaps already 50% minimum proficiant??).
So it takes the top computers in the whole world to become even better and faster for 10 years so far and even more years to be able to drive a car as safely as a human. That's why the US Government recently announced that they are coming after Elon and Tesla with legal action for it's false advertising and claims about the Tesla FSD full self-driving mode, because it can't self drive safely. You can't blame the FSD mode when you get into an accident and damage property or injure someone. That is always the driver's fault.
This super computer would blow up the AI industry. Processing speeds could rise exponentially. We have also progressed in terms of storage. Imagine AIs having these types of sources to produce better results aside from learning portion of the neural network. I've been seeing Bluewillow training their AI by inviting testers so I can tell that the future of AIs will be bright indeed.
actually big part of dojo is the ability to do parallel training. meaning that it can feed the data to all of the tpus at once instead of breaking the problem up. this allows to speed up training which would be impossible on regular tpus no matter how many of them you have
@@princeofexcess Both GPUs and TPUs can be used for ML. The idea is that TPUs can be better optimized because they deal with matrices natively. I don't know which would be better for extremely sparse matrices, but it would be an interesting question.
You should take a look at Alethea AI. They are introducing CharacterGPT. We can create interactive AI characters by simply entering some text. Also they are working on the ownership of AI generative content.
Lets hope they could equipted them not just w/ visual sensor but w/ athere sensored just for a reserved or suport as an auciliary suport system so it would give it more sufficient and safer travels.
I just heard last week that The US government is preparing to take legal action against Telsa for its FSD mode and claims. There are RU-vid videos with the FSD Telsa drivers sleeping at the wheel. They should get a major conviction for dangerous driving charge by the police, because the robot driving the car isn't advanced enough to drive a vehicle safety by itself.
If they are using dojo to speed up ai for FSD, they can apply the same method for many other human capabilities and expand their ai. Over time, it will be huge.
6:00 actually thats not exact. tesla gains data from every car at every moment wether people have bought fsd on their car or not. It is always running in shadow mode.
It would seem that the “input” is the most critical part. Then you can process it. What are they using now. LiDAR, radar, cameras, ultrasound. A combination of it. The simpler the input, the easier on the back end.
The future is rapidly moving towards becoming fossil fuel free. The Neome project in Saudi Arabia is an example of a linear city that doesn’t have cars. I believe there’s a future for automobiles in linear cities although, there will be far fewer vehicles for automotive transportation at that time. There will be a great need for vehicles like buses, gravity ramps, maglev, trains, wheeled trains, bicycles, elevators and glide walks. We will have many kinds of transit paths in this future. This is why I believe a stronger future for AI will involve data mining to solve problems that are too difficult for people to find in the mountains of data that we can accumulate.
Sounded good and people unfamiliar seem to be impressed. However, you need to go back and do some more research on the reasoning for the Not Secret Dojo Project and the Not new AI concept.
When Apple couldn't obtain the chips they desired from vendors, they took matters into their own hands and created the most powerful laptop CPU chip in the world, yet it went largely unnoticed. The same situation occurred with Tesla, who sought to create a more efficient supercomputer CPU chip, but this went largely unnoticed until the stock of Nvidia, their main supplier, suddenly plummeted due to Tesla becoming a competitor in this market segment.
5:00 "...they're likely retraining the network on a weekly basis...": And look at the competition. Driver's ed takes a 15 year old and trains him to operate a car by the time he is 16. Actually, in my case that training was less than 3 months (70 years ago). And the training (at that time) paid no attention to the most important input (i.e. situational awareness). I wonder if Tesla's AI is paying such attention. If it was, in the previous clip, the car should have been proceeding at a crawl... that wasn't the only door that might have opened. A real "usefull" AI would have refused to let him drive on that road... and fired all government workers who had anything to do with its design.
@@carholic-sz3qv talk with ChatGPT and ask it. The news always hypes stuff up like this. ChatGPT has limits the creators made to it can’t make any human like or sentient anything like that. It basically just gives you answers to like simple questions
But we're no smarter than the yeast used to brew alcohol - yeast keeps making alcohol till the percentage of alcohol gets high enough to kill the yeast. Scientists - including Exon's own, have been telling us for decades that burning FF is causing global warming which causes climate change and we're still drilling for more.
I remember rumors about TESLA car CPUs being able to mine crypto when resources aren't needed. Is that really a coming thing? Wouldn't it be more advantageous for TESLA to integrate the computing power of the millions of their cars to contribute to their computing power? Wound't that decentralize it?
It makes me wonder if a super computer can identify what needs to be learned in a year, and what needs to be unlearned in a year to discern between the two facets of what is happening and what is not happening over time. Especially with new construction on roads with detours and bridges. And when that work is finished to erase that info to free up more memory capacity. I truly doubt a super computer has the capacity to operate a car over the life of a human being. Plus I wonder if Organic Neurons travel at the same speed as AI Neurons.
I really like how he charges people $15,000 for beta testing his FSD feature... after promising everyone they could make $30,000 a year with remote taxi service... LOL anyone who bought this has a class action lawsuit for lost wages every year they had the thing
4:00 "...reducing the errors...": Interesting clip. Tesla was directing the driver into the left lane on the other side of the pillars (a badly designed roadway) ...just as a guy opens his door into the driver's lane. Tesla immediately removes the directive. But it doesn't stop the vehicle. And the screen is useless to the driver (actually worse than useless). If he had been watching it he would have been unnecessarily distracted and there's no telling what he would have done. How can they be this far into AI and have the screen where it is? At least we're now starting to see it moved to the driver's line of vision (i.e. Audi). But why don't they have heads up displays? What we're really seeing is "AS": Artificial Stupidity.
