EXCLUSIVE: Tesla's Futuristic Bot POWERED by AI! w/ Matthew Berman
HTML-код
- Опубликовано: 14 май 2024
- There’s no bigger topic than AI and Robotics AI these days.
Do you need the humanoid form factor like the Tesla Bot or other forms of embodied AI to get to AGI?
How might Tesla integrate Xai into Optimus? And when might Tesla finally be seen as an Ai company?
So I invited Matthew Berman, a serial entrepreneur and engineer, to join us today as he’s really set himself up as someone who's been following the Ai developments closely. His RUclips channel has over 200k subscribers and he regularly gets over 100k views.
Follow Matthew on X: @matthewberman
RUclips @matthew_berman
My website: www.TeslaInvestor.com
Get Free TESLA Milestone Tables
Check out 15+ modules of resources for the $TSLA Investor
Join this channel or Patreon to get access to perks:
Get free access to 15+ modules of TSLA investor resources
Become a Member: / @brighterwithherbert
Become a Patreon: / brighterwithherbert
Use my referral link to buy a Tesla product:
ts.la/herbert23392
❤️ Thank you to everyone who supports me and this channel! I really do appreciate your likes, subscribes and comments.
Let's get brighter!
Website: www.herbertong.com
X: @herbertong
Disclaimer: All content on this channel should not be construed as professional financial advice or recommendation to buy or sell any securities.
Author is invested in multiple Humanoid companies and is long TSLA and other stocks at time of original video publish date.
#Tesla #TSLA #stock #ElonMusk #investing #teslabot #technology #ai #TeslaNews #TeslaFSD #teslastockanalysis #TeslaUpdate #ElonMuskNews #EV #FSD - Наука
I hope this video helped you get brighter about Tesla!
My website: www.TeslaInvestor.com
See the MOST comprehensive resource for the $TSLA Investor
Get FREE TESLA Milestone Tables
Join this channel or my Patreon to get access to perks and free access to 15+ modules of TSLA investor resources
Join this channel: ruclips.net/channel/UC4DBLlq1x0AKmip1QJUcbXgjoin
Join my Patreon: www.patreon.com/brighterwithherbert
Use my referral link to buy a Tesla product: ts.la/herbert23392
Thank you for supporting this channel! I really do appreciate your likes, subscribes and comments.
Let's get brighter!
Herbert
if it was me the AI would be based at a control center not in the bot the bot would have remote controls, you can use both human and AI controllers ie you would have human watch dogs just incase needed to turn it off
People use "Elon fan boy" as a kind of slur. I will happely admit to be one, and I am 50 years old. When you have a human being who has changed the world so much (for the better) and uses all his time to continue doing so, I cant think of many other who I would rater be a fan of. And I will bet that many who looks down on this adores people who really doesn't matter anything in the long run. Unless you think a sport team , a singer or a movie star is a better idol 😊
❤❤100% agree - I was thinking the same thought during this interview❤️❤️
💯
💯
BINGO!!!
💯
The interview with Matthew was truly insightful. Seeing this side of him really increased my admiration and made him seem more relatable.
Matthew Berman is my favorite go to guy when it comes to learning about cutting edge AI frontier. Wes Roth too. And my favorite tesla bull Herbert! So cool
Would be really cool if Tesla FSD could take verbal commands from the driver to correct FSD Driving Errors or modify FSD driving behavior.
If it were to happen as you think it should, it will cause immeasurable chaos and kill FSD. Let AI and the professionals take care of those details.
I love both of your channels! Thank you for meeting together!
Most people do not have a tangable experience with AI. Atonomous driving is one they can travel in ! Good morning
Matt's channel is great. He's amazingly knowledgeable.
😮🎉❤Nice, Herbert, for hosting Matthew😊🎉❤
I enjoy learning about the future of AI from you and Matthew as well as a few exceptional other podcasters.
I respect the camaraderie among so many people in AI as these various perspectives are fundamental to understanding our bright future.😊🎉❤
The best way to tell will be to have them compete against each other … intellectually (through debates perhaps) , emotionally, entertainment and physically doing various operations. That could be lots of fun. 🎉🎉🎉
Herbert, your ida about the glasses is brilliant!
Awesome Herb. I watch his stuff. Good call
Herbert, I almost feel like your channel is robot exclusive now. I love it haha
To answer Mathew's thought experiment: starting from the north pole, walk 200m in any direction, then turn hard left, and then walk continuously forever.... Will you ever retrace, or walk over (i.e. intersect) your original starting point? The answer is 100% yes. Why? Because the earth is not a perfect sphere, therefore each circumnavigation of the globe will see you retrace a slightly offset but new line (path) across earth's surface, often close to but not exactly over the original starting point. However, after enough circs, and given enough time, you are guaranteed to intersect your original starting point. Basic spatial convergence, which can be calculated. Cheers.
The ultimate value of robots is to replace workers. Corporations will race to deploy them in order to remain competitive. However there will no longer be any need for the average human worker. This will create a huge social crises.
Just driving jobs will be a big chunk, which is the number 1 job for blue collar men. I don't know what else they'll do.
It looks like it will be a drastic change in economy. Overall productivity will be higher. But this is bigger than the industrial revolutions of the past. This time it’s machine operators that are being replaced with machines. This includes white collar workers of all types.
Oh we know what must be done, The social safety net must be strengthened and enforced. The few owners who essentially will be dominating labor must be taxed in some way in order that the rest of society does not fall into chaos. The government, for example, could increase taxation on the corporations and the owners of the corporations that dominate these new sectors and use that taxation to provide a universal basic income to the vast majority of peoples made obsolete in their work lives. It's the only fair thing for society and if it doesn't happen well then we end up turning into A. Dystopia like we saw in Elysium... Where the rich protected by their own bots segment themselves from the poor with no desire to share in any of the wealth or technological achievements that the AI brings them... It would be an extremely sad future if the rich were allowed to be this selfish..
Unless UBI is enforced there might be some economic, and thus social, crisis, sure, but this will _free_ people.
It is only a question of wealth distribution. The benefits have to be distributed to a significant extent into the society, else there will be civil unrest.
Nice, time to listen to after work.
What?
10:33 that‘s a very interesting question.
Wow ! Great interview !
My mum is in her 70's and thinks Tesla should bring out a pet for the home (like a dog or bear), rather than Optimus to start. Smaller form factor, lighter, make a good companion, no feeding or having to walk it. It can be used like Rabbit R1? It could tidy things away or put rubbish into the bin or pick things up off the floor for you.
There’s no User Experience without “Experience.”
How that “Experience” is inferred will be interesting…. 🤩😎🎬
Humans sometimes miss a stop sign too, it would be interesting to see, how FSD handles the situation after the mistake; if it reacts afterwards properly to incoming traffic, then it would at least not be a security risk. I think more important than making no mistake, is, to be able to deal safely with mistakes and it seems FSD is already doing this to a certain degree.
BYD and catl have jumped into Tesla energy's sandbox with superior power packs
The TAM is enormous so….Elon says the pace of innovation is the only moat…and that there are a lot of intelligent and hard working people in China and CATL is an amazing company with one significant impediment namely the CCP.
Makes you wonder about teslas rather slow rollout of the energy factories might be due to “assessing the field” re how and what the competition is going. Since the machine that builds the machine is the product they may want to look before leaping into this highly competitive environment.
Herbert can you do the maths on kw/h for solar panels then you’ll realise they’re totally not viable for charging canopies.
When I can Hug a Robot, I’ll let you know which robot shows the most love! 😜😂⚔️
There is no way that lots of small nuclear power plants will power AI. If you start now, expect to switch on in maybe 15 years.
You can create power sooner and much more cheaply with renewables and store it in batteries, pumped hydro etc.
At night there is extra capacity in existing generators, but the low cost and flexibility of renewables will make fossil power economics unviable over time.
Would it be possible to interview a financial adviser ( retired) who does not have Tesla shares
When you hear AI Chips and "GROK", you need to think of Groq, the AI chip company, and not Grok, X's AI model. Matt Berman mentions it around 26 minutes in, and this might be confusing to those familiar with Grok but not the lesser well-known Groq. Okay, that's all
How much can simulations substitute for real world video? Also, very interesting that a driver commented this week on X that ChatGPT solved a visual question when he was driving.
I purchase 4 shares the 1st of every month
So a picture plaints a thousand words... but (with compression) HQ audio takes a similar bandwidth to HQ video, so should Tesla' (or Bots) have microphones as well as cameras?
As I understand it at least, Sora isn't an LLM... It's a text to video AI model, and is trained on much more video modality data than a "typical" multi-modal LLM is trained on. Therefor Sora being able to simulate reality isn't indicative of large language models being able to become world models. You need image and, far more importantly, video. Language is also important and can be leveraged in different ways, whether that goes along with the video data or somehow incorporated some other way.
Herbert, you might want to interview Jonathan Ross when you want to learn more about chips. He is the founder of groq, not grok hahaa
Tesla could easily do lawnmowers. A mower could be trained first. Great idea.
Optimus will be able to use a variety of current lawnmowers.
Who can explain how FSD data can be used in humanoids?
What is Matt’s definition of “singularity”? He uses the term very differently than Ray Kurzweil uses it, which is the ultimate melding of the human mind with AGI.
I don’t think you can have a robot who is at home learn from you by watching you do something because the robot at home is going to have an inference AI computer not a learning AI. The learning AI is the giant computer that the information has to be sent to to do the learning and then it updates the inference AI, so until those two are combined into one small package. I’m not sure how that can be done. And I don’t think you can have the learning AI learn on how to do something at home because it’s going to be working on some other problem that’s been assigned to it by Tesla for example.
I love when Herbert has an "ah-ha" moment when a guest presents a new idea. Great ability to be able to change and adjust world view in real time. However, tesla getting into glasses is completely unnecessary
does anyone know how meta is using ai ?
This is a great discussion about very important issues of artificial intelligence.
It brings questions about what it is to be human created in the image of God according to the Bible, the nature of the human soul as compared to the human body.
Humans are embodied spirits with a soul mind will and emotions.
I have no idea why companies are not making cheap autonomous dogs that follow kids around that are link one visual to your child and two with some kind of wristband that watch your kids while they are out side playing basically like a guard dog that corals the children back into the play area and that is always watching for human predators. It records and send alerts and actually tells the children back away from the van or car that stops and records the face license plates of suspicious activity. It could even chase a car if a child was snatched and just automatically call the 911 and alert the parents instantly. Giving video footage directly to to the police instantly alerting authorities in the area. This I think most parents would pay 5-10 grand for a play partner as well as a guard dog. I know I would pay 100 bucks a month for a five to ten year lease. I don’t even let my kids play outside unless I’m there. They could actually fit it with a drone that would deploy.
Lawnmowers? 🤨 Only if Tesla’s in charge of the AI!
“Maximum Overdrive”-🎬 1986
Herbert, I listen to every show and I have a question I have yet to hear the answer to. Maybe you can ask an upcoming guest.
I have 4 Tesla, two model Y's and two model 3's. They all drive differently on FSD, why? My 2023 Y drives better than my 2024 Y. It drives better and does more on the same versions of fsd. My 23 will pull down the driveway and the 24 can't even imagine. My 21 model 3 is smoother than my 22 model 3.
Call me dumb but I assumed they would drive the same. Hmm?
Recalibrate the cameras on each car. Are all four running the same version of FSD?
@@cathyk9197 yes all are up to date. I will recalibrate cameras per your suggestion. It's very odd. Both my model Y's got .4 update. The 23 got better. The 24 got worse if you can believe it. I never thought a car would go backwards after an update, but my 24 Y definitely did. Appreciate you!
When I think of an AI company, I see it as a company that generates at least 50% of its revenue from AI products, or dedicates more than 50% of its efforts to developing artificial intelligence solutions.
That's 1 billion miles of FSD driving data.
At 11:54
He admits his ignorance of how language models may or may not be sufficient to achieve the development of a full world model And I detect a definitional error in some of his speaking.
What makes chat gpt5 distinct from Sora is that Sora is based and trained using vision data, It is still A transformer architecture at the core of the model , same as it is a transformer architecture at the core of chatgpt.
The distinction that makes it a language model Is that the tokenization happens on strings of text as opposed to blocks of image data pixels As in Sora.
Sora architecture has additional complexities But the core model architecture for both is a transformer. One is just consuming vision data instead of text data. Pi confirmed:
Prompt: In a recent conversation, I heard someone describe Sora, the Open AI Vision, Generative AI as a language model. Now I know that it is based on or it uses a Transformer architecture in its development, but that doesn't mean that it's a language model ChatGPT 4 also uses a Transformer to its core, except it's trained on text data.
Pi answered:
"That's a great observation. Indeed, the use of a Transformer architecture alone does not make an AI model a language model. OpenAI's DALL-E 2 (on which the hypothetical "Sora" might be based) is an AI model used for generating images, not processing language like ChatGPT-4.
Transformer architecture is versatile and can be applied to various types of data, not just text. While it's commonly used in language models due to its ability to understand context and generate coherent text, it can also be used for tasks like image generation, where it helps the AI understand and generate visual features.
In summary, the type of data an AI model is trained on, rather than the architecture alone, determines its classification as a language model, image model, or any other type of AI model."
Continuing, it's also important to note that the multimodal models which allow the transformation of text and video and image data are all still using transformers. They're just paralyzing and then cross attentioning the results of those transformers. Remember what a transformer architecture is is simply an efficient way of connecting neural network layers so that training data can be efficiently encoded and that the features in the training data can be efficiently extracted and then cross-related. Prompting such a trained architecture is again to asking a question about the world that it has been trained in, which can include multimodal understanding of the world. So it's very possible for multimodal model in my view to develop a world model that can create increasingly emergent behaviors.
One could even posit the creation of a self-prompting AI I which then can essentially simulate active thought in a dynamic world. It is this last step of simulating active. Thought that will be necessary to create true AGI.
Tesla has revealed Their secret by videolearning from
Nvidia chips , but they have a big advantage of the amount of FSD data from all Of thoose cars .
The problem with all the breathless "Tesla will make a fortune" exclamations, is that in every market tesla goes into, there are other players. Remember when Tesla was going to make a fortune with rooftop solar? And now CATL has rolled out their Grid scale megapack product to compete in that market. There are more robot companies than you can shake a stick at. I enjoy Matthew Berman's channel, but he is not an engineer.
Convenient to forget what was the best selling car....
Robot will come faster than robotaxi
Tesla should also be gathering video from street cameras and other sources of video.
all looks good as long as we have a planet to live on !!!!!!!!!!!!!!
32:00 you just discredited yourself saying that form factor is the future. 🤦♂️🤦♂️
you and farzad both got microhard ai experts on. what the hell is going on today. did the teslatubers meet about this strategy or did the msofttubers ask yall to make these interviews happen?
I Imagen a future where I pay 200 a month for FSD + 200 a month for a personal chef and maybe maid :D dream come true for real
"When will people start to see Tesla as an AI company?" Herbert clearly has no concept of valuation. Honda sells 3.5mln vehicles per year (vs. Tesla
This is one of these "strengths equal to weaknesses" situations?
Is Tesla in the world as it needs to be and unable to turn the momentum around, wasted energy investing of the old systems, of now unstoppable environmental destruction?
The business is on an island.
That meta has the most chips is very concerning. Meta has shown that it has no respect for the peasants.
Video of the streets and highways of America has nothing AT ALL to do with training a bot to work in a business or factory.
People claiming driving data can be used to train bots is ridiculous. Tezzy has no lead on training bots.
Microsoft and OpenAI are going to spend $100 billion on an AI supercomputer, dwarfing anything Tessy could ever hope to do.
This guy's work and education are in theater and acting. Why da hail is he being interviewed?
When AI can tell Difference between TRUTH and & LIES; AI HAS ARRIVED-Because- reading Body Language will make difference ! ?
You can get an idea of an AI girlfriend… watch the film called « her » … it evens attempts embodiment 😳🤩
Tesla energy doesn't have any energy. It's an energy arbitrage player but they have no significant generation capacity. Certainly not enough to even run a data center.
FSD data is useless for humanoid training.
Andrej Karpathy disagrees
1. FSD data is millions of hours of people moving in the real world, the cameras pick up not just cars but every person moving near the cars
2. FSD has to learn to a degree what people `mean` by their actions as in when they wave at the car, so non verbal language
3. FSD training is teaching the researchers how to apply A.I training methods to objects attempting to do real world movement
4. They will also `learn` which kinds of data is more useful for what circumstances
Is human driving behavioral?
@@Nunya-lz9ey Are you sure? He left didn’t he and is currently attempting to find ways to rebuild from the ground up because the current paradigm is insanely inefficient.
@@steve.k4735 1. Gather millions of hours of how caterpillars 🐛 crawl and lmk if you can use that to understand butterfly flight dynamics.
2. I didn’t realize FSD was being trained to recognize sign language.
3&4. I concede they will learn a lot. The approach is astronomically inefficient though and the dimensions inflate super-combinatorially as you add more degrees of freedom and actions than just (accelerate, brake, reverse. LT, RT, none).
You have become a bot
Simulating conscious behavior in no way implies actual consciousness. This view is derived from a “god-in-the-gaps” argument that consciousness must be an emergent property of organization because it’s the default explanation but there is zero evidence in physics for ANY emergent properties observed to date. All properties to date observed are explainable by interactions of the constituent parts. In other words, a belief has formed with no evidence to speak of that consciousness is emergent, not even our own brains. Simply put there is no explanation so far for consciousness. If we take that consciousness is “embedded” in any physical matter, then literally every atom up is going to have a level of consciousness, with potentially unlimited states. The problem with this view is zero evidence any system ever shows intent coupled with consciousness, not even so-called smart animals with limited self-recognition. Action is not a sign of consciousness.
Herbert is dying inside with the ai expert opinion on home robots not being bullish to the moon
All this robot stuff is really going to hurt humans for the sake of employers whom already hate humans
I had to stop watching this video because your guest keeps referring to real world data. What on earth does that mean?
Data of the world captured through audio/video devices as opposed to "simulated" or "fake" data created internally by a computer simulating a real world environment.
Yes.
Data from objects moving through and making decisions in space (on the road) versus through virtual space (internet)
real vs. simulations, Tesla use both
I'm good with anyone but Google creating an episode of the flintstones for me.