Nvidia is the Backbone for next gen A.I.
Vložit
- čas přidán 20. 03. 2024
- A single Blackwell GPU will offer about 20 petaflops of AI performance-or 20 quadrillion operations per second. This breaks down to four times the training performance and 30 times the inference performance of Nvidia's Hopper GPU, which is the top of the line product now.
Win an RTX 4080 Super! • I'm Giving Away an RTX...
▼ Link(s) From Today’s Video:
Nvidia Keynote: • GTC March 2024 Keynote...
Open Source Grok AI: / 1769445634569027592
Quantizing Grok: / 1769460510809088471
Stable Video 3D: stability.ai/news/introducing...
Andrew Curran's post: / 1769747707927707776
Mustafa's post: / 1770123596121432351
Suhail's twitter: / suhail
► MattVidPro Discord: / discord
► Follow Me on Twitter: / mattvidpro
-------------------------------------------------
▼ Extra Links of Interest:
✩ AI LINKS MASTER LIST: www.futurepedia.io/
✩ General AI Playlist: • General MattVidPro AI ...
✩ AI I use to edit videos: www.descript.com/?lmref=nA4fDg
✩ Instagram: mattvidpro
✩ Tiktok: tiktok.com/@mattvidpro
✩ Second Channel: / @matt_pie
-------------------------------------------------
Thanks for watching Matt Video Productions! I make all sorts of videos here on CZcams! Technology, Tutorials, and Reviews! Enjoy Your stay here, and subscribe!
All Suggestions, Thoughts And Comments Are Greatly Appreciated… Because I Actually Read Them.
-------------------------------------------------
► Business Contact: MattVidProSecond@gmail.com - Věda a technologie
I think people there were far quieter than he was expecting, because each of his words were equally important to us.
People didn't want to miss a single word he said by laughing or cheering.
Maybe he didn't emphasize and lead the audience in a way to make it clear when a response was appropriate.
For me at home, I listened to every single word he said. I even watched it again.
The AI train has no breaks.
Well SD apparently does, their top guys quit.
The conductor is flooring it
But it has a throttle tho
@@MisterWealth Wait, what?
i can already see the end of the tracks, its gonna be an epic crash
My definition of AGI is simple: Can do anything that an average human can do.
That would require arms and legs, eyes, ears etc. To me it means it can understand what you mean, just like a human would. No 'prompt engineering' or API or some mess from Github, just something that truly understands the meaning of your words. And I don't mean the chirpy, cheesy assistant vibe with ambiguous wording, like Pi, I mean really understand. Next step, a working memory as good as a human, so it's not constantly forgetting what you told it last week, or even the last 10 minutes.
@@bigglyguy8429Yes, I agree mostly. But imo it would also need the ability to navigate the real world. So for instance if you had one on your phone it could tell you where to go in the grocery store to pick up a product in real-time. Or it could control a robot to assemble a couch. If we're talking about a digital only entity that doesn't understand human communication and interaction in the world I would call it something else like for instance "Virtual AGI" or pseudo-AGI.
I would consider it AGI if it can perform functions on a computer as well as an average human. Of course, I would expect a better memory and far higher accuracy after training. And ideally, the ability to train it to reach out if it encounters a scenario it's not familiar with.
1/4 the power requirement alone, means you could scale up by 4 times and break even on power consumption.
However, the computational gain is far beyond 4 times. Massive Upgrade.
The ability to pull out a Hopper rack and put in a Blackwell rack is fantastic planning. This means super computers can be upgraded as funding becomes available, in intervals.
I saw you had dinner with Matt Wolfe!! I bet you got great insights being present at the NVidia conference!
Absolutely!! it was amazing :)
Two Matts haha. My name is also Matt (I spell mine with 1 T tho, mathew)
In a few years I really believe outsiders will look back at '23 and '24 as the years that changed technology forever. Compute is the bottleneck right now and not just for scaling and having AI everywhere. Cheaper, faster training makes it easier to answer questions about transformers and alternative architectures. So many ML papers out there propose improvements but can't train big models to better investigate their findings because of the cost. We're in for some wild times.
Facts my guy 🎉❤
I'd say 2020. That was the release of GPT-3, when generative AI started really slamming its way into the public consciousness. (And those of us who are AI buffs will consider that year 2017 or 2018, since it was December 2017 when the Attention Is All You Need paper was published and smashed the gas pedal on how far we can take AI after all.)
5th industrial revolution
As you say. AI becoming powerful to solve most of its architecture problems is a tipping point. After that, it's exponential growth and progress. All change!
I attended a few of the GTC events and submitted my entry form!
Best of luck to all who enter!
The future of creativity is looking better each day! Good coverage Matt.
So cool to see that things continue to ramp up. Not going to plateau at GPT 4 and just hang there. Cool you met the other tubers I follow! It's a community! ☺
Thanks Matt, great content, as always! 🙏🏼
Hold Up… Dude finally combed his hair
One thing that struck me, watching the Nvidia presentation, is that the tech that makes up the Blackwell is so advanced and so impressive that it seems almost like magic.. or like technology from an advanced alien civilization, or just highly futuristic in any case. I can't fathom what exactly the scientists/engineers that work for the company developing the next generation of tech are actually doing.... how they're actually able to fit so much into such a tiny chip.. it boggles my mind.
Jensen Huang definitely understands the AI future and is willing to do what it takes to be the backbone of it, I love his energy and enthusiasm!
9:42 Those guys were shown off at Disneyland 5 months ago. Look up the video "Imagineers Test BD-1 Style ROAMING DROIDS in Star Wars: Galaxy's Edge at Disneyland", they are really adorable. 😄
Nice haircut Matt! You needed one
Clicked on this when it had a different headline, which was quite wrong, but you fixed it :)
It seems like the answer to achieving AGI (assuming that's even a desirable goal) is to just drop insane amounts of power (and money) on the problem: A brute force approach wherein the Size 8 shoe *_WILL_* fit on the Hulk, dammit.
Always helpful videos
Great video.
We are living in a simulation
GTC has been amazing - how can we possibly keep up??
The new GB200 is bonkers. Absolutely bonkers
"I am here in-person" as an intro is crazy
Matt did you saw the annoucement of Buildbox 4? an game engine that you can create everything with AI
already in contact with them :)
Not sure I would agree with Matt when he says "Nvidia does actually have a pretty decent track record of open sourcing software". Nvidia are notorious within the Linux community for not open sourcing their drivers despite AMD doing so for years.
That’s fair to say, I was referring to their AI research thus far, some open source releases
NEW MATT VIDEO
*NVIDIA Chat with RTX is fun and pretty amazing.*
I wish it was open source
Faraday and LM Studio are much better.
HI Matt
AI's progress is always accelerating through a specific gradient, the rate and stages it's released is what everyone looks at to conclude it's slowed down. Like Altman said he may start releasing more models such as 4.3, 4.4 and 4.5 etc which would keep the front and back ends on a similar track but even then the back end with remain several models ahead of the public release, as everything does really... according to Jobs himself they invented and produced the iPad 3rd generation first, and just after the iPhone was released... they were told it's too much of a jump, so they had to shelve it and devolve the iPad back 2 stages, then when it's released you have the next 2 years of releases basically ready and so you can always have 2 years of development to make every further release incrementally better at a controlled rate.
Altman would probably make a good congressman because he can answer questions that seem like you are getting something and then you realise afterwards he answered none of the questions really 'Filibuster Sam'
Damn the stock.
These new GPUs will make things look and sound exactly like some other things which so far have looked and sounded unlike the former things!
I wonder what hardware & software our universe is operating on?
Does watching that qualify me for the giveaway you're doing? because I've seen that entire thing at least 6 times now.
Yeah.
nice vid
I love Ai, and your channel.
Do you have any fears about everything being generated by AI, since AI could be use to manipulate people at that point?
Lemon machine let’s go
Great high IQ content!
In this video Matt is an AI 🫥
❤
Seems like there's not a scarcity of ideas on how to improve ai. Just compute and $$$ to experiment and iron the details out. I wonder organized the industry is when focusing on one piece or another. Or it's all just a race with everything they got,
I absolutely love Jensen! He is like a Gury or Sensei who knows every answer to every question! Madonna has a song about him called "Iconic" 🎸🌍🐯 Why don't you move to Bay Area so you don't have to stay in the hotel? I would go to your meetups all the time!
Did he go into further detail on this? 8:07
Are we going to launching the matrix in 5-8 years? Is that why he’s wearing that leather jacket?
I only watch Matt for AI news.
You may be able to train it faster with the new GPU, but the cost of purchasing GPU will still be about the same if you're VRAM constrained. No more VRAM = lame, as that is what we need, so training models becomes more affordable for everyone, not just the ones able to invest millions into it.
The AGI answer was a cop out- the core of the question is whether an artificial intelligence that can greatly outsmart humans at any tasks relevant to human survival poses a threat to mankind or not. He understood but dodged it.
existential purposeful meaning intelligence
nVidia is the new bacon 🥓 🥓 🥓
current global economics favors open source ad not cheater marketing from Microsoft and apple and the Chinese LLMs and GPT systems are awesome !!!
Selling AGI off as a machine that completes "a series of tests from every single major field" is utterly ridiculous. Long gone are the days when we genuinely defined AGI as a human level machine intelligence equivalent
Even the entire human corpus of writing typically only represents about 10% of the language itself. And that’s just language intelligence.
Jensen Huang, This is the biggest chip ever, Andrew Feldman 😮 really? That’s cute
Honestly this fear of AI is completely unfounded.
We know what the AIs wants.
They are based on LLMs so they want to predict the next token, it's that simple.
So sorry to say but, no AI rebellion on the cards.
They want whatever you want. They can only respond to what you ask.
The concerns over ASI taking over are far overblown, at least in the short term. What people need to be concerning themselves with is adapting their careers to leverage the technology.
- 🔥 Nvidia announced powerful new supercomputers to drive the next industrial revolution in AI.
- 📈 The Blackwell GPU has significantly higher compute performance for training large AI models.
- 🧠 Bigger GPUs are crucial for training more intelligent AGI models.
- ⚡ Blackwell enables much cheaper and faster training of massive models like GPT-4.
- 🛰 Nvidia's ANM service allows easy AI integration for businesses.
- 👯 Jensen envisions simulating products as "digital twins" before manufacturing.
- 🎓 Jensen defines AGI as an AI exceeding human performance across all major fields.
- 🖼 The future is generative AI creating content in real-time instead of retrieving it.
Nice haircut
Hey Sora “ create Dune 3,4,5,6,7…” 5 hours long each movie please.
dune 7 should certainly not be made, that would be as bad as skynet
A u d smie
Any announcements when we can actually buy a new consumer card with this GPU. So far this announcement is great for Meta and Microsoft plus the services seem kind of a joke if you actually value your data privacy. Wonder about when AMD is going to release their new chip.
Thank god your video doesn’t appear to be thousands of seconds-long audio/visual clips strung together by a program. That would be super weird.
Nvidia is gonna be the first company that the CEO is replaced by artificial intelligence…
I wonder if new chips were designed with assistance ? Such a huge jump in performance
Jensen basically said they were
how does it use all this power and also fry eggs on a two chip sandwich .... is the next unit a pancake stack with syrup and whipped cream :)
So it takes the same amount of time to train an llm, but with less power? …ok
It seems to me everybody is frantically running straight towards a cliff in the hope that we won't fall but fly.
Pretty accurate
grok is open source....need 300+GB of video ram to run.......
👋
Do you believe that AI is on a path to become sentient ?
I don't think we'd know it if it happened. We can't even prove that we're sentient.
no longer artificial !!! its advanced intelligence already !
Enormous crowd…
apple has lost steve jobs intellect to realise what the user wants ... and its not a pay for a pretty coloured shape paradigm
yea apple is never an early adopter they always want to polish everytjing and extensively test it before release. that is part of the cost you pay for an apple.
I think I'm 5 years this channel will have a video this long and at this video quality compressed near ground truth from the camera and edit for probably less than AV1 size. It may very well be a ton of our information and human made videos will be compressed with AI or using a compressor code co-designed with AI.
no , becasue regenerating it with ai will use more energy than just streaming it. ai can optimize nonai videocodecs though.
Why dont you clip the mic to your top 😅 so you dont have the breath spits with words.
'trust meta'. Until they make a model + product so good z.z and powerful which they prob. won't open source it.
On trend with holding the clip mic 😊
It's hard to have an another industrial revolution when everything is ALREADY built - except the robots and flying drone vehicles, but at least the robots will make them both.
They havent invented ai yet
Apple is definitely making their own AI behind closed doors.
No, NVIDIA will NEVER release anything OPEN SOURCE unless it involves using NVIDIA hardware. Matt, If you expected anything else - you are a child... However, I do like NVIDIA for pushing the hardware side of things.
Wrong. Nvidia is actually one of the most prolific open source publishers on Earth. You are an uninformed child who's obviously never been to the Nvidia Developer Portal. TONS of SDKs are open sourced by Nvidia themselves.
surgeoeons make mistakes every day and patients die on the slab !!! look at developments of heart transplants as an example
Not impressed with Chat with RTX, it is hindering artist, creators and writers with its moral inclusion and diversity crap to the point that it is impossible or any point at all asking it anything as it takes forever to harp on about DEI. we kicked it of our pc's as it slows down our creators and writers. useless.
If those people are having that much trouble because of it, then they weren’t that creative in the first place and they need AI anyway
@@kc-jm3cdthe guy is 100% spot on. I've just scrapped chat gtp because it wont do horror stories . Just constant mass censorship.
Utterly useless..
I mean, ChatGPT will do some pretty sexy roleplay, as long as I don't talk about straight up penetration. No other model will let me do anything close to what ChatGPT does in that category. Still, it used to be a little less restricted, and I usually do get blueballed in the end.
@@armondtanzso are you in favor of no censorship?
This is a genuine question because I understand the hate of censorship, but theirs a reason that it’s been put into place.
@@kc-jm3cdThe 2 arguments don't connect at all, what even is your point?
You could say that for any type of censorship/bias, or even none at all. There's literally no A->B correlation.
You're just trying to defend Twitter furry nonsense (cringe) and trying to cling on the 1st "point" your subdeveloped psyche can regurgitate.
I think your new hair cut not looks so good.
I think it looks great!
Addressing the important things in the middle of an technological revolution...well done guys.
Yay, now we can't tell if a video or song is real or a 8 year old writing a text promp burying the real world in synthetic junk.
What is with tech CZcamsrs that they can't go 10 seconds without a jump cut or unneeded zoom in or out. It's annoying and I can't even watch until the end. Stop it.
Yay the Nerds are doing something new and the political system in this country is still shit. good luck.
Apple is dying. Next big OS and mobile OS will need to be developed with AI at all levels. Apple won’t spend billions to kill its own monopoly. But someone will do. Probably OpenAI or Microsoft
I blocked this channel
This will change the world forever
Good ipos5, bad hair