and reading this article about the importance of AI technology disseminating in all corners of society and the economy. The potential benefits of AI technology are vast, from making people smarter and more creative to automating tasks and jobs that can lead to greater efficiency and productivity.
The comparison to the invention of the printing press, which made people more literate and informed, is apt when considering the impact that AI technology can have on society. Just as the printing press revolutionized communication and knowledge sharing, AI technology has the potential to enhance human intelligence and capabilities in ways we cannot yet imagine.
The push for open-source AI models, as advocated by leaders like Mark Zuckerberg and Yan Loon, is crucial for ensuring that AI technology remains accessible and transparent. By allowing developers and researchers to collaborate and build upon existing models, we can accelerate progress and innovation in the field of AI.
The debate over the regulation of AI research and development is ongoing, with concerns about monopolies and control over powerful AI systems. However, the emphasis on open-source models and democratizing access to AI technology can help mitigate these risks and ensure that AI benefits everyone, not just a select few.
As we continue to explore the potential of AI technology and its impact on society, it is important to consider the role of openness, collaboration, and transparency in shaping the future of AI. With the right approach, AI technology has the power to transform our world for the better and unlock new possibilities for human creativity and innovation.
its funny Meta is lost 😀 just copying whatever they can.. missing all original thinking.
Yann also says nearly daily on Twitter that LLM'S have zero intelligence and are doomed to fail because of confabulations, which cannot be completely trained out. You always get exponential degradation of output accuracy with these models.
I'm not sure why Mark is not looking more into Cerebras for a weighted steaming option. I believe it's the way forward to feed the insatiable hunger of LLMs
At about 7:30 "It turns out that coding is actually really important structurally for having the LLMs be able to understand the rigor and hierarchical structure of knowledge, and just more generally have an intuitive sense of logic." in this quote you can also swap out LLMs with Humans, and it will be broadly very correct based on what I anecdotally have seen and heard and experienced myself as a now quite senior computer engineer. Learning to code was a step function in my cognitive toolbox, and gave me a much deeper intuitive understanding of mathematics.
Other courses I took along the way and also their application in production has given a much better intuition for both inductive and deductive logic as well as systems thinking and a deeper appreciation of scale. Being able to think about code from single instructions executing on the order of nanoseconds to cache hits or misses in tens of nanoseconds, memory access around 100 nanoseconds, blocks of code running in microseconds, HDD/SSD IO in milliseconds, network calls from single to tens or hundreds of milliseconds, and then data processing running various O(N) scaling laws on 1 to thousands to millions to billions of pieces of data need to be structured and done very differently and at each level of scale or abstraction you have different concerns and implications.
This has also helped me appreciate and get a bit into geopolitics and macroeconomics, where the sheer scale of things like demographics over time and the size of supply chains in the economies in various industries can be mindboggling if you don't understand scale. Like to complete the green revolution by 2050 will either require AGI/ASI take-off with robotics covered fairly soon (looks potentially plausible), as without that there's not enough capital available and getting produced to pull off the build-out of energy and manufacturing infrastructure as well as skilled labour to man that building and manufacturing in time to meet the challenge.
With AGI being predominantely developed via being open source, that means AGI will be the first technology achieved via humanity coming together. The normal everyday person can contribute from anywhere in the world. That's awesome.
A rule I learned a long time ago, is, no matter how stupid, controversial, or even plain wrong idea you present, there will always be the 3% who are for it. Always.
Any proposition that gets less than 3% naysayers, has to actually be super valid.
P=NP=AGI
Meta/Zuckerberg should want Jann LeCun to stop talking. He is very unlikeable, and when future AI disasters occur, he will be a reason Meta takes a lot of blame. (Or maybe they just feel invincible after having survived the 2016 election fiasco…)
Excellent commentary on open source AI and power of sharing and learning from open source, democratic and free AI. My own experimentation using multiple AI models together as open source has proven very useful.
How can the Chinese companies buy H100's? Aren't they under the tech embargo?
That pencil neck Mark
weights are the source code the knowledge base the neural network.
1:35 Glad to hear Meta guy arguing for decentralized power of AI, but can’t help being skeptical given Zuckerberg’s czar-like rule over social media and the attention economy for 4 Billion users .. an unelected life-long reign, even though we can see he’s not good at this powerful job
No question you could come up with is equivalent in terms of Gods knowledge. GPU's use lots of gold on the chips though, so I could see why they're hoarding the gpus.
The GPUs contain precious materials. That's why they hoard the gpus.
Why are we selling advanced chips to China? Have we forgotten they're a Communist dictatorship? Have we learned nothing from our mistake of enriching China *because greed* so they could modernize and become an existential threat to the whole world? 🤦🤷
Thx.
Stockpiling GPUs is a bad strategy for figuring out AGI, not even close.
On "everyone will have their own AI enhancing their intelligence", it comes to mind that repeatedly throughout history people have tried to liberate themselves by concentrating power. Whether it's through the government, or AI, these often do not end well. Something big and powerful enough to provide for all your desired needs, is also a monster that can swallow you whole.
Someone will control the AI you rely on, someone with their own agendas and priorities.
I don't believe any open source model that is more powerful than gpt 4 will be released.
Funny how OpenAI is actually "closed".
11:58 WERE IS THIS POLL?? I was literally talking about exactly that topic to everyone with an ear today.
That guy, went to number 1 spot regarding AI personalities.
I still do not know enough about the guy to trust him, but the Zuckerberg turns out to be a voice in favour of privacy as the giant truly empowers the open source space. LLama 3, going open source… if it is multimodal with an appreciative performance in text mode…… you have two braincells.
Open Source the Singularity!
Yeah, who needs investment in housing, health care, education, roads, living wages, and pensions? I'm sure homeless folks living in tents can't wait to get their own personal AI assistant.
I would have never expected someone high up in a major corporation to be for Open Source, since they're priorities are usually:
1 – Projecting the company's personal morals, and sacrificing a percentage of paying customers to do so. (censorship)
2 – Money. (keeping everything closed source to stay ahead of other companies)
3 – Technological Advancement.
Them saying they plan on open sourcing AGI leads me to believe the number one priority this time is technological advancement, which is definitely a curveball! With Meta's resources, I believe they can advance really fast by leveraging how Open Source works!
He might be but from everything i heard i dont think so
These corporations have a god-complex.
after trying to get ChatGPT to generate an image that it deemed "inappropriate" (the MIDDLE FINGER OOOHHH HOW OFFENSIVE), I am now of the opinion that open source is mandatory so we aren't constrained by the moral puritans of some corporation.
No matter how far Closed Source gets, as long as they continue enforcing absurd levels of censorship, Open Source will always win.
If Zuck has majority control, is that why Elon wants more shares? To be able to effectively compete with Meta? Or just a billionaire thing?
What is open source worth when it rewrites its code based on proprietary training data?
I don't want AI controled by my government either here in Germany.Our government sucks.
I appreciate that you point out the author's opinions, and offering a more neutral and fact based approach when talking about it. This is excellent journalism.
Limiting it to H100s is probably hiding the reality of the investment level and compute capacity of those companies since several are building their own chips.
Also, I thought Chinese companies were restricted to H800’s.
I really like all your content and what you offer. The thing that really bothers me though is your editing with the closed caption yellow follow the words part. You only used to do it a tiny bit in your earlier videos for emphasis. Now it's just way too much and it's like having your CAPS ON FOR EVERYTHING!!! Please stop!
i don't understand, why people view next more advanced Ai much more dumber than current one? It's illogical nonsense, more smarter Ai cannot make evil by intelligence.
Open source AI is a deceptive term. In the traditional sense is when the source code of some software is accessible to the general public. Neural networks have an additional element to their structure that contribute to the production of the software, the training data. With AI we can open source the weights, and the network's architecture code while still having a proprietary element with the data used to produce the model. With Llama, as well as other models, claim to be open source in the traditional sense but is not truly open source because the data used to produce the weights are hidden. I really think that one of the most important things to do for regulator purposes, and safety is to have an open source architecture, weights, and data set. I really feel that it is really misleading to call Llama, Mixtral, and other models open source when the production of the software contains an integral element that is not open and accessible to the general public.
Fake News 🗞️🤯
If meta is the first to achieve agi I will eat a shoe
Zuck has become an expert at blowing money. It’s only a matter of time until Facebook goes out of business.
Any AI get this question right?
have 5 balloons today, I popped 2 yesterday, how many do I have?
Can someone explain where the benefit is for Meta to go fully open source with LLAMA?
I think there wont be much usage of the Accelerated datacenters. You need paying customers. I see demand for normal compute but demand for accelerated compute is not there right now
brother I'm not gonna lie, the captions that are like lyrics to a song piss my off, very distracting
Bad guys always find ways abuse system while good guys respect rules. So it might be safer as open source.
Now that engineering has met philosophy, we have a new religion. Will this group try to force us into compliance like the Catholics or Muslims or not sees before them? They already are.