@Jerryvhall1@Hypercycle_AI HyperCycle is an enabler of the internet of AI. The clients of HyperCycle owning the 600,000+ Node Factories are producing nodes that enable any and all AI agents to be plugged into the internet of AI. I sometimes call global brain
Hey @saylor , as you're gearing up to stack more BTC, remember: the night is darkest before the dawn, but with Bitcoin, we're always on the brink of a new day. Keep HODLing strong! #Bitcoin#HODL
@marinetondelier Interdire un réseau social comme Twitter/X revient à attaquer la liberté d'expression, même si ses dérives sont problématiques. La solution réside dans l'éducation, la régulation intelligente et la responsabilité des utilisateurs, pas dans une interdiction. #Liberté
Yes clearly we have not achieved Human-Level AGI yet in the sense in which we meant the term when we published the book "Artificial General Intelligence" in 2005, or organized the first AGI Workshop in 2006 or the first AGI Conference in 2008 ... the things that put the term on the map in the AI research community...
What was meant there was not merely having a generality of knowledge and capability similar to that of a typical humans (and to be clear o3 isn't there yet, it's way superhuman in some ways and badly subhuman in others), but also having a human-like ability to generalIZE from experience to very different situations... and no LLM-centered system I've seen comes remotely close to this. I have not had a chance to play with o3 so I can't say for sure but I would bet a lot that it still has similar limitations to its predecessors in this regard.
Modern LLM-centric systems come by their generality of knowledge and capability by a very interesting sort of learning which involves -- loosely speaking -- extrapolating a fairly small distance from a rather large volume of information. Human-like AGI involves some of this learning too, but ALSO involves different kinds of learning, such as the ability to sometimes effectively cognitively leap a much longer distance from a teeny amount of information.
This more radical sort of "generalization out of the historical distribution" seems to be (according to a lot of mathematical learning theory and cog sci etc. etc.) tied in with our ability to make and use abstractions, in ways that current transformer NNs don't do...
Exactly how far one can get in practice WITHOUT this kind of radical generalization ability, isn't clear. Can AI systems take over 90% of the economy without being able to generalize at the human level? 99% I don't know. But even if so, that doesn't mean this sort of economic capability comprises human-level AGI, in the sense that the term AGI has historically been used.
(It's a bit -- though not exactly -- like the difference between the ability to invent Salvador Dali's painting style, and the ability to copy Salvador Dali's painting style in a cheap, fast, flexible way. The fact that the latter may be even more lucrative than the former doesn't make it the same thing.... Economics is not actually the ultimate arbiter of meaning...)
About the AGI-ARC test, when Chollet presented it at our AGI-24 event at UW in Seattle in August, I pointed out after his talk that it clearly is only necessary and not sufficient for HLAGI. What I said is (paraphrasing) it was fairly easy to see how some sort of very clever puzzle-solving AI system that still fell far short of HLAGI could pass his test. He said (again paraphrasing), yeah, sure, it's just the first in a series of tests, we will make more and more difficult ones. This all made sense.
I think o3 model kicking ass (though not quite at human level) on the first AGI-ARC test is really interesting and important ... but I also think it's unfortunate that the naming of the test has led naive onlookers and savvy marketeers to twist o3's genuine and possibly profound success into something even more than it is. It appears o3 is already in real life a quite genuine and fantastic advance. There is no need to twist it into even more than it is. Something even more and better will come along soon enough !!
I have found @GaryMarcus 's dissection of the specifics of o3's achievement regarding AGI-ARC interesting and clarifying, but I still find what o3 has done impressive...
Unlike @GaryMarcus , I come close to agreeing with @sama 's optimism about the potential nearness of the advent of real HLAGI ... but with important differences...
1) I somewhat doubt we will get to HLAGI in 2025, but getting there in the next 3-4 years seems highly plausible to me.... Looking at my own projects if things go really really well sometime in 2026 could happen... but such projects are certainly hard to predict in detail...
2) I don't think we need to redefine the goalposts to get there.... I think automating the global economy with AI and achieving HLAGI are two separate, though closely coupled, things... either one could precede the other by some number of years depending on various factors...
3) I don't think the system that gets us to HLAGI is going to be a "transformer + chain of thought" thingie, though it may have something along these lines as a significant component. I continue to believe that one needs systems doing a far greater amount of abstraction (and then judicious goal-oriented and self-organizing manipulation of abstractions) than this sort of system can do.
4) However I do think transformers can provide massive acceleration to AGI progress via serving as components of hybrid architectures, providing information feeds and control guidance and serving many other roles in relation to other architecture components.... So I do think all this progress by OpenAI and others is quite AGI-relevant even though these transformer-centric systems are not going to be the path to AGI unto themselves in a simple way...
5) I think it will be for the best if the breakthrough to HLAGI is not made by closed corporate parties with "Open" in their name, but by actual open decentralized networks with participatory governance and coordination... which is how all my own AGI-oriented work is being done...
@SingularityNET@OpenCog@ASI_Alliance
@bengoertzel@KalyanRH How do you see Hypercycle's (and other companies) avant-garde approach shaping the integration of decentralized AI agents in the economy, and what unique advantages could it bring to this rapidly evolving landscape?
It has been a busy week for setting up Node Factories🔥
3 Masternodes🎆
1/2 a Masternode🎇
Multiple smaller level Node Factories set up💪
Over 2 million HyPC liberated onto mainnet removing them from circulation 🚀
Set up your Node Factory and join the global AI economy! 🌍
It has been a busy week for setting up Node Factories🔥
3 Masternodes🎆
1/2 a Masternode🎇
Multiple smaller level Node Factories set up💪
Over 2 million HyPC liberated onto mainnet removing them from circulation 🚀
Set up your Node Factory and join the global AI economy! 🌍
Marc Benioff states that AI agents are doing a great job, raising doubts about hiring prospects in 2025. A bold vision highlighting AI’s growing role in business. What’s your take on the future of work? #AI#FutureOfWork#Salesforce#AIAgents
Salesforce CEO Marc Benioff says the company may not hire any new software engineers in 2025 because of the incredible productivity gains from AI agents
💡 Decentralized Computing = A Game Changer!
Say goodbye to centralized control and hello to global collaboration. #NuNet bridges diverse hardware and software into a seamless, decentralized compute network.
🔗 Learn more: https://t.co/41gxommGkH
Here is a very specifically defined research programme aimed at creating deep neural networks (including transformer-like nets with additional features) that understand compositionality and are able to incorporate genuine reasoning. This is distinct from my work on symbolic and evolutionary AI ... it is a proposal regarding how to better create the "neural" side of neural-symbolic AI.
Step 1 is to get rid of backpropagation in favor of predictive coding and information geometry; then Step 2 is to approximate and guide this information geometry based on hypervectors embodying concept algebra (incorporating compositionality naturally).
There are many details to be worked out by experiment, and this will take a while to explore.
https://t.co/IRMW8EJnkB
My colleagues and I will explore this via open source code and if it works will deploy it via decentralized networks under broad democratic control, focused heavily on applications of broad benefit to humanity.
@True_AGI@SingularityNET@ASI_Alliance@nunet_global
AI pioneer Ben Goertzel has just revealed a groundbreaking research program aimed at creating neural networks capable of understanding compositionality and integrating true reasoning.
I disagree with the anti-Musk narrative pushed by some journalists like Aurore Lalucq, and here's why: innovation. Elon Musk has transformed industries with Tesla and SpaceX.
He's made electric vehicles accessible to the masses and pioneered space travel for commercial use, achievements that were once considered sci-fi. Critics like Lalucq focus on Musk's social media presence or business decisions, but they often overlook his groundbreaking contributions to technology and the environment.