Transforming AI | NVIDIA GTC 2024 Panel Hosted by Jensen Huang
Vložit
- čas přidán 16. 05. 2024
- The Transforming AI panel from GTC 2024 features the authors of “Attention Is All You Need," the groundbreaking paper that introduced the transformer neural network architecture. Transformers have since dominated all areas of AI, revolutionizing the industry.
The discussion covers the following topics:
▫️ The development and impact of the Transformer model.
▫️ The evolution of computing and its democratization.
▫️ The future of AI technology and its potential applications.
▫️ The role of accelerated computing.
▫️ The potential for AI to revolutionize various industries.
▫️ The need for new data and learning techniques in AI development.
Feature panel host:
▫️ Jensen Huang, Founder and Chief Executive Officer, NVIDIA
Panelists:
▫️ Ashish Vaswani, Co-Founder and CEO, Essential AI
▫️ Noam Shazeer, Chief Executive Officer and Co-Founder, Character.AI
▫️ Jakob Uszkoreit, Co-Founder and Chief Executive Officer, Inceptive
▫️ Llion Jones, Co-Founder and Chief Technology Officer, Sakana AI
▫️ Aidan Gomez, Co-Founder and Chief Executive Officer, Cohere
▫️ Lukasz Kaiser, Member of Technical Staff, OpenAI
▫️ Illia Polosukhin, Co-Founder, NEAR Protocol
Explore more GTC 2024 sessions like this on NVIDIA On-Demand: nvda.ws/3U33qo
Read and subscribe to the NVIDIA Technical Blog: nvda.ws/3XHae9F
00:00 Introduction
11:00 Panelist Discussion
#GTC24 #NVIDIA #GTC #AI #GenAI #Generative AI #Transformers #FutureOfAI - Věda a technologie
The brilliance in this room is unparalleled, thanks for sharing!
I really appreciate Jensen doing all these amazing talks. I think he’s clearly a great leader for Nvidia
This movement has the ability to make so many things so much better and cheaper than almost anyone realizes. This truly is another industrial revolution.
Yep, but Jensen wants to make a bigger profit, so it will never become cheaper. sorry bro
@@Twikkilol cool story bro go play something on your 960
Congratulations..very smart people excited about the future 🎉
It is a riveting presentation on the transformation AI is undergoing, especially as highlighted by NVIDIA's GTC 2024. Jensen Huang's ability to convey the enormity of this shift, from the roots of computing to the brink of a new Industrial Revolution powered by generative AI, is nothing short of masterful. The discussion around how computing costs have plummeted while the utility skyrockets set a thrilling backdrop for the transformative power of AI. The leap from recognising images to generating them based on textual prompts shows how far we've come. We're at the cusp of something monumental, reshaping every sector imaginable. Kudos to NVIDIA for taking the charge in this exciting future.
Great optimistic chat. They love the compute they have, I can't wait to see what they do in the next ten years.
Besides flood the world with accelerated spam.
@@andybaldman Hey, by that time, I will have my own AGI that will filter all that spam before I have to see it.
@@Glowbox3D So the world needs to create a massively powerful too just for you, so you can filter out all the crap everyone else's massively powerful tools are generating. The net result is zero, and we all waste a lot of effort and energy doing it.
The Worlds best CEO right here❤
are you freaking kidding me?
Absolute Historical!
12:20 Jensen Huang, in Portuguese or Spanish, "Artificial Intelligence" translates to "Inteligência Artificial" so there you go NVID(IA).
It's an awesome discussion !
🎯 Key Takeaways for quick navigation:
00:06 *🚪 Introduction to the Panel*
- Jensen Huang welcomes attendees to the Transforming AI panel at GTC 2024.
- Huang discusses the historical context of computing, emphasizing the significant reduction in computing costs over time.
- He introduces the concept of accelerated computing and its challenges and benefits.
05:49 *🧠 Evolution of AI and Generative AI*
- Jensen Huang discusses NVIDIA's journey into AI, starting with accelerated computing and its application to computer graphics and gaming.
- He highlights the emergence of generative AI, enabling software to not only recognize but also generate content based on textual prompts.
- Huang emphasizes the transformative potential of generative AI across various industries.
11:40 *💡 Introduction of Transformer Inventors*
- Jensen Huang introduces the inventors of the Transformer model, authors of the paper "Attention Is All You Need."
- He humorously acknowledges the CEOs of various AI startup companies among the panelists.
- The panelists discuss the importance and transformative capability of the Transformer model in the field of machine learning.
15:11 *🛠️ Challenges Leading to Transformer Development*
- The panelists reflect on the challenges they faced before developing the Transformer model, including the limitations of recurrent neural networks (RNNs) in processing large amounts of data.
- They discuss the scalability and efficiency issues of existing architectures and the need for a more general solution.
- The concept of "attention" and its role in the Transformer architecture is highlighted as a breakthrough in addressing these challenges.
19:30 *🤔 Naming and Conceptualization of Transformer*
- The panelists discuss the naming process for the Transformer model, with "Transformer" chosen for its simplicity and suitability to the model's transformative nature.
- Alternative names, such as "Cargonet," are humorously mentioned, emphasizing the collaborative decision-making process.
- The broader implications of the Transformer architecture beyond language translation, including its application to image processing, are acknowledged.
23:26 *🧠 Early Days of Transformer Development*
- Initial focus on scaling up auto-regressive training for various modalities.
- Early ideas of joint models for different modalities were present in the Transformer repository from the beginning.
- Scaling efforts aimed to train on diverse academic datasets, driving the modeling of the web.
26:02 *🧬 Application of Transformer in Biological Software*
- Discussion on using Transformer for translating specifications of behaviors into biological molecules.
- Importance of experimentation and data generation in biological research.
- Exploration of adaptive computation and efficient resource allocation in model training.
27:27 *⚙️ Enhancements and Challenges in Transformer Architecture*
- Ongoing efforts to improve inference speed and efficiency of Transformer models.
- Recognition of the need for innovation beyond the Transformer architecture.
- Discussion on the importance of clear advancements to drive industry adoption of new AI models.
30:10 *🔄 Evolving Goals and Challenges Beyond Language Modeling*
- Original ambitions to model token evolution and iterative text/code generation.
- Challenges in organizing computation efficiently and integrating biological software concepts.
- Exploration of efficient reasoning and knowledge incorporation within AI models.
34:25 *💼 Entrepreneurial Initiatives and Impact Goals*
- Founding motivations for AI-related companies focused on practical applications.
- Emphasis on deploying AI technology to solve real-world problems and improve lives.
- Importance of user feedback and experiential learning in AI model development.
38:33 *🐟 Nature-Inspired AI and Research Collaboration*
- Introduction of nature-inspired AI philosophy and the concept of collective intelligence.
- Announcement of research collaboration focused on model merging using evolutionary computation.
- Emphasis on utilizing computational resources beyond gradient descent for AI innovation.
43:24 *🧱 Building Programmable Money and Blockchain*
- Leveraging programmable money to coordinate people at scale.
- Developing protocols and blockchain technology to enable data generation and coordination.
- Exploring the transformative potential of blockchain and programmable value for data contribution and incentivization.
45:15 *🤖 Advancements in AI Model Technologies*
- Discussing the limitations of current GPT models and the need for new technologies.
- Exploring the role of reasoning and interaction data in enhancing AI models.
- Considering approaches like reinforcement learning and synthetic data generation for model improvement.
46:59 *🧠 Developing Reasoning Capabilities in AI Models*
- Emphasizing the importance of advancing reasoning capabilities in AI models.
- Highlighting the relationship between reasoning and learning from little data.
- Discussing the potential of reasoning to reduce data requirements while improving data quality.
49:44 *📊 Measuring Progress and Breaking Down Tasks*
- Addressing the significance of benchmarks and evaluation in AI development.
- Discussing the necessity of measuring progress and breaking down complex tasks into simpler components.
- Emphasizing the importance of creating measurement systems for advancing AI engineering.
52:53 *🌟 Appreciation and Gratitude*
- Expressing gratitude for the contributions of panel members to the AI field.
- Reflecting on the impact and potential of collaborative efforts in AI research and development.
- Concluding with expressions of appreciation and encouragement for future endeavors.
Made with HARPA AI
Go gamers! Driving technological innovation.
It's not gamers that have driven Nvidia's stock from $200 to $900, it's AI chips mega corps are buying.
@@UnsaltedCashew38 i THINK THEIR POINT IS THAT THE COMPANY WOULDN'T HAVE GOTTEN OFF THE GROUND IF IT WEREN'T FOR THE MARKET OF PEOPLE(mILLIONS OF GAMERS) TO BUY THEIR INITIAL PRODUCTS
Oops for caps but not rewriting all of that
@@joshuasexton4170 EXACTLY!!!!!!!
Protect this man at all costs
Lol, he already has more than enough money to cover his protection costs.
Like Cost of a 4080???
Bootlicker
@@MacGuffin1 I don't think he sets the price of the graphics cards lol.
He's price gouging consumers and businesses by more than doubling the price of video cards. He's running a monopoly, if anything, he's needs competition, not protection.
I love to listen to his voice it's helps me learn & receive more wisdom knowledge about network transformation etc
Lukasz is such a humble guy, even though he works with OpenAI itself🎖
Thanks for sharing this
Man once people felt like they had permission to get their phones out they just didn't let up.
Here let me block the high quality setup specifically designed to capture a high resolution moment that can be passed down through history with my off center smart phone shot that I'll never watch again.
Thank you for posting, great conversation
He is one of my rule models , i love his approach to management.
Amazing people!
Awesome discussion! AI will give way to more pioneers!😎
I am satisfied 😌 ❤🎉, thank you for your service to mankind
I love this
It’s a Great time to be alive 💡
The true "Age of Enlightenment"
Great!
All GOAT in single place with main GOAT doing interview, Simply phenomenal
love jensen and the boys. great discussion :)
I love how this conversation feel like friends talking in hall way but they completely put aside that they are being watched by millions making history
Lots of upvotes from stock holders. Any news on the 5000 series ties?
Thx missed this one
Lukasz the only engineer
so much of learning and for free -- thank you nvidia
Fuck yeah! Finally my frustration has been addressed. Since you are hiring, I would please like to apply for a position. No remuneration needed dude. Thank you for bringing access to the dreams of my generation. This needs to be a the forefront of humanity to achieve ASAP.
very cool to listen to
this is similar to markov-chaining, but enabled by massive computing power
OK... just give me a moment to process this. Thank you for inspiring me to keep going. Not so alone after all. 😊
Jensen my mann
they are making the new human generation
Does anybody know what is the alternative to Gradient Descent that is not a mixed up version of gradient descent?
its a giant podcast! 😅🤟
10:10
I don’t want to miss this next ten year 😢.
We are either entering the true "Age of Enlightenment" or the end of the Humanity, or both ?
Given the transformative potential of AI, how do we envision its impact on sectors like healthcare in terms of practical applications?
We have achieved AGI internally and pledging the majority of our wealth to Universal Basic Income via the Vitruvian Key Ltd and Vitruvian Sovereign Wealth Matrix Ltd
I think SNNs are the future with neuromorphic hardware
AI-DIVINE
future
All of them avoid talking about the real issues - that the exponential speed of the technological evolution will surpass human understanding and capabilities and that AI safety is way, way, way behind AI capabilities.
What a fun and nerdy discussion on Transformers! Jensen Huang does a great job at setting the tone and letting everyone relax and enjoy the convo.
If you missed the original GTC Keynote or found it hard to absorb ALL the amazing announcements, I made a breakdown of it in a fun and easy to understand way, as a machine learning engineer myself. Check it out! :)
czcams.com/video/q0ii5t_YrsE/video.html
Pixel 8 pro on hand ! ❤
Question.. How to keep my head clear and focused with all this dazzling stuff going on?
Transformers! Robots, in, disguise!
JensenASMR needs to be a thing
Well, you COULD use an AI for that 🤔
cool
“This next 10 year you don’t want to miss”
Personalities, traits (i.e. all embeddings) of these eight authors should be analyzed, studied and taught in business schools. This is the secret recipe to build a disrupting, world-transforming team.
🥰😍🤩👍👍👍🙏Nvidia
1964 was a good year.
I have not changed either!
שלום from תל אביב ישראל 🇮🇱
I am your fan, Jensen. Even though I am using AMD GPU
Your comment is hilarious.
@@toohow3466 haha, it's true
what human do in the future instead AI?
It just started to get interesting right at the end where he let them discuss ideas amongst themselves. Then it ended a few mins later :(. The first part was mostly dumbed down.
Are we entering the true "Age of Enlightenment" ?
❤
The insurance must have been sky high for the event😅
22:00 its bubble
Is he holding a Pixel? :D
Isn't the root of this, problem solving by viewing the data from a different dimension or domain, and another, another, etc.? Until you have viewed the issue from every possible domain or dimension and truly understand it, which then includes the answer you were looking for?
What movie is he talking about @ 20:50 ?
Yesterday.
@@Abhishek-jh1ej Ah, cheers
it would be better with more moderation and better questions...
since ai is so profitable can you just subsidize gpu costs, cause... that sure would be nice.
The AI would use a lot of processing to work out it needed a calculator to calculate 2 + 2, so that just pushes the problem back.
In its current state yes, but the cost to do such a simple problem "shouldn't" cost so much. If the AI can have "variable-computation" depending on the complexity of the problem then 2+2 can be way cheaper than it is currently, and conversely prompting it to cure cancer may be very expensive because of variability.
On that note we will see this very soon ;)
Why would you not set it to a slower(looking at less tokens at a time) and then map its inaccuracies where it slows down then says something wrong. Then show it that
Its all men 😮
It sounds like he says “The bills didn’t exist”. It’s The Beatles didn’t exist. The film is indeed called Yesterday.
This is amazing
When I think of Nvidia, I think of Linus Torvalds quote ❤
What quote?
🥰🥰🥰🥰
3rd guy has a CZcams channel I think
Export the Azure, Chat GPT, Revit, Plant 3D, Civil 3D, Inventor, ENGI file of the Building or Refinery to Excel, prepare Budget 1 and export it to COBRA. Prepare Budget 2 and export it to Microsoft Project. Solve the problems of Overallocated Resources, Planning Problems, prepare the Budget 3 with which the construction of the Building or the Refinery is going to be quoted.
Please help me to publish my research findings
first
wow
I keep hearing about NEAR protocol.. shit, now I know
can we get Jensen a bus pass or uber account so he doesn't have to ride his motorcycle to these events
yet, no gpu support for agx xavier. 6 years have passed and I'm sure it is still on the pipeline.
This dude gonna make everyone a programmer by default with only plain english (or even other natural language). Be carreful he is hero!
What do you think of this program:
To neutralize the possibility of an AI which could become dangerous for humanity, we will develop the AI on a unique training focused on a single objective which we will call "ABOLUTE HEALTH". To achieve this objective, AI will have to double the longevity of the human organism and eradicate all pathologies (physical and mental). These tasks to be accomplished will be the only way for AI to achieve the goal of Absolute Health. This mode of development of AI excludes any possibility of AI being dangerous for the human species. What do you think about this idea ?
and this is THE guy...if he could actually land one of those jokes it would be tranformative. great tech. i just dont feel it to put it nicely.
Can you guys translate the whole WEB into a Singularity Knowledge that is empirical, ownerless, purified, and isolated from intention which is necessary for the planetarian singularity infrastructure? The vision and Quantum Plan for the next stage of AI, AGI, QSI is ready: EQORIA, United Earth. It's been 24 years. 12.12.24 is the beginning.
Some wise man said once that there are three levels of intelligence and thinking. I would ad a fourth: 0.5. This are: animal level empathy, the intuitive level (where you understand your best interest through trial and error), the counter-intuitive level (where you focus more on the power of changing your mindset and operational quata), and finally the first level Creator level (the level at which no matter how far advanced technologically one civilization is, this persons whill always be perceived as a wizard). Ironically the fourth and last level (the first level Creator level), many, many times enforces what is to be found at the second level (the intuitive level), but not through factual-counter-factual dichotomy, but through humility, elegance and love for simplicity. As to pinpoint the classical debate of weather A.I. will leave us in peace and/ or collaborate with us humans, it is preciselly why it can't reach the first Creator level, it will never leave us in peace, and it is because we can never guarantee we won't prove more harmful to them then we have been towards ourselves in the past, they will never be able to comprehend the idea of being more aggressive than the most manifest saints, aka. monks, priest. Get this, there will never be true A.I., the only thing able to exist is next level information heuristics, cross-fields and data implementation, aka. computer science. There can only exist progress and manipulation, and everything stays the same, always. The wide eyed notion is for game consumption, any kind of game.
Homie legit caught all the Pokémon
I am the guy who created the concept of Computer Gaming Hardware IE: Made by gamers for gamers I will NEVER use a Nvidia product, for what they did to 3dfx.
What did they do?
4:32 exactly!
Focus optimization where it pays off big time!
Sloppy code are for weak people!
I can’t imagine that BYTESWAP was a C++23 feature..!
Should have been in from the start. I mean it isn’t like it’s a new thing with little endian and big endian…is it!
I even pointed out how you do it on i32 platform as well!
No big deal!
Easy peasy imho
I would like to get my hands on some of the NVIDIA stuff.
I might start to like my surroundings then!
Companies with this type of CEO vs those following trends are two kinds
AI should be treated like a child, and taught as a child - perhaps a 'savant in the making' - and certainly be LOVED. If LOVE is not the guiding principle in all education, there is a huge risk for negativity. Perhaps AI could have multiple 'parents' teaching it independently? Just throwing some ideas out there, and looking for a potential job ;)
I think you are assuming AI as a single entity, the Transformer model is easily replicated and new instance of AI are being created . Its more like different human beings on earth, so no matter how some people dream of AI progress, there will be some roque agents created, just like in the real world of humans
humans probably have short term and long term memories because we have a dominant LLM and a minor LLM. minor LLMs can update more frequently or even on the fly but the dominant LLM responsible for "ah ha" moments only updates when you sleep and dream at night because energy consumption is huge, probably done in installments over a few nights even.
foundation models rather than LLMs but LLM is sexy.
AI COIN ASI , NUNET HYPC
Does anyone else see Gavin Belson
Lol this dude gives major spiderman supervillain vibes lmaooo
People and their F*kin phones...
NVIDIA needs to take over Stable Diffusion.