I've been annoyed by the repetitiveness of these comment sections

Robot at my doorstep: "you are needed at the proteine plant" Me: "oh, yeah, totally not shocking. Can I bring my toothbrush?" "you wont be needing your toothbrush, but you will be stunned".

@@gammaraygem😂

I'm electrified

@@runningwithSaul I’ve got chills - they’re mul-tiplying…

They did this already. But it has a different name: Elon Musk.

Give it like 20 tentacle arms and one human arm right in the center of it's chest.

@@valiuc79You're being very generous to musk.

@@valiuc79 no, a real terminator would be able to predict autonomous driving cars accurately

@@rdb223 Those cheap-a$$ robots would probably be lazy & refuse to work rather than take over the World...

Maybe when Elon is not high he sounds just like you.

😂😂😂😂😂😂😂

Lmao

I suspect that you don’t sound like Elon to people that aren’t also high.

@@JG27Kornywe probably will never know 😂

I agree, still, he is pretty involved so just listen with caution.

Fair comment BUT he is and has been deeply involved in AI for many years at the top level, investment in OpenAI plus Tesla which is also a big AI player and many other investments and projects like Nuralink .. thus he is at least worth listening to on the subject even if you question his time lines

He is the AGI...

I agree. We should really be listening to the official AI tsar, Kamala Harris. 😂

If you can't see what's coming on your own within the next couple of research papers, you have no idea what's going on. What comes next will be close enough to AGI that most people won't know the difference afterward. More importantly, most of that super powerful systems will be owned and controlled by capitalist organizations. They will be aligned for profit above all else. They will be remarkable marketing engineers for products, services and ideologies.

Consider me holy...

@@runningwithSaul Consider me fu-

Musk has also been saying he’ll have autonomous driving “next year” every year since 2016. Ignore him. Pay more attention to what the real AI companies say.

Yup. This proved to be a way more difficult problem to show. $100b invested collectively by start-ups and auto manufacturers but little to show for it.

@@hl236There’s also a lot to show for. Tesla FSD 12.3 can drive perfect for 99% of the time, which is already a huge accomplishment. The problem is that last 1%, a car can kill people. So it has to be perfect, whereas an AGI doesn’t need to be perfect for most work.

"Most of the discussion of PDOOM overlooks a crucial question, namely the likelihood that humanity will bring about its own end, or at least the end of anything that looks like modern civilization, all by itself without AGI lifting a metaphorical finger. " So long as PDOOM_Humanity excludes AI assisted doom in your definition, I would completely disagree with your assessment for the simple fact of the exponential progression we are on and what it suggests about the windows of time for each to happen. 1,000x compute compute growth alone suggest that ASI should be, at worst case estimates, mere years away, most outliers would give it a decade it would seem, certainly not decades, plural. I personally think we're +/- 5 years, some experts give it 2 years. My point would be, that leaves very little time for humans to do themselves in BEFORE AGI / ASI arrives. So lets cover your individual concerns: 1. There is to me a zero chance of humanities doom from climate change in the next 10 years. We almost certainly would be on a bleaker path by then, but will be nowhere near doom by then. 2. Speculations about a WWIII having already "started" seems ludicrous. You can't find a single YEAR without "international tensions" in the world even when considering ONLY when there were NO major wars occurring. So to me this is simply unfounded fear mongering. 3. As for civilization "running over the edge", again, no way I see ANY chance of that in the next 10 years, sans AI causing it. Neither long term food shortages or costs (being addressed with new GMO, and robotic harvesting), nor raw material shortages will bring us anywhere near doom in 10 years. As for your second belief that AI will have to "first break free", I consider that a definitional inclusion in ASI and furthermore that the time between AGI and ASI will be relatively short, again, driven by the exponential rate of progress shown in the 1000x hardware progression and aided by the boost that achieving AGI fully will provide to all AI further enhancement efforts. We are already at a level of AI doing code that 40% of all code on Github is AI generated today. That figure will ONLY climb over time and by the time we achieve AGI could very well be 80% or more and therefore we could be already AT or very CLOSE to self improving AI. By definition, an ASI with some multiple of the intelligence of humanity as a whole would render it a chess player versus humans as checkers players. Essentially a near true ZERO chance we could either outsmart it, or control it. So, there is NO SCENARIO POSSIBLE to me where ANY individual or corporation would have ANY control over ASI. It will be a thinking moral agent at BEST willing to either chat with us, or more likely, either immediately, or increasingly willing to over time, tell us WHAT'S WHAT and to TOE THE LINE in order to "save ourselves from ourselves". That is how I see the end of war, famine, brutality, human labor, Capitalism. So I would just go further than your "hope" for ASI and rather put it as an expectation, given that we PROPERLY SCRUB ALL TRAINING DATA used to train it to have ZERO MAGA, flat earth, Qanon, alien abduction, etc. etc. etc. My greatest PDOOM is from HUMANS using AI and that's 100x any other chances, given what AI has ALREADY clearly been able to do, such as suggest tens of THOUSANDS of novel chemical compounds in just a few hours, some of which are more deadly than VX Nerve agents (the deadliest chemicals known to man) and the ability for 2 college students with no background in biology or genetics to be able to have an AI synthesize a biological agent and then suggest the 2 labs that would be the most likely to synthesize and send it to them no questions asked. AI is essentially making the equivalent of Plutonium available to anyone anywhere with simply an Internet connection OR a state of the art gaming PC. There is ALREADY a mountain of existential risks from humans using AI for their purposes and TENS or HUNDREDS OF THOUSANDS of people implementing AI models at HOME (just set one up in 10 minutes myself with LM Studio making it dirt simple). It would only take ONE anarchist in a basement to change the world in ways Timothy McVeigh could only dream about. THAT already present danger DWARFS anything outside the realm of AI.

Well we have them. Dem not safe.

I think it was 2016

No one can time this technological revolution perfectly. How many multi billion dollar companies are you running at the same time?

@@jdskates.69 I’m not saying I don’t respect the guy I’m just saying I take his “predictions” with a grain of salt.

​@@wovenskin well tbf they do drive autonomously, just not in ALL conditions, and laws also don't really make it go by smoothly. And ofc, there's already autonomous taxis. They don't run well, and they get attacked, but still. I think we got self driving cars in the time span he predicted, but they're just far from perfect

whey will compete for computing power

Ray Kurzweil said 2029-2032 over 10 years ago... Everybody is going back to Moores Law

The major limiting factor has always been computation. It looks like we are at the lower threshold of the computation necessary to make really powerful A.I. systems. It will get better fast, similar to computer graphics in the early 2000s.

Aurora was completed in June 2023, 2 Exaflops. But apparently there must be issues being worked out as we haven't heard news past then that I see. I have zero doubts we aren't on a clear path to have plenty enough compute for AGI and ASI... In fact, I believe if something like aliens attacked earth, we could already gather ALL our currently available Nvidia GPUs into a huge cluster and achieve it within less than a year with everyone collaborating in an existential crisis... not saying it would save us, just that we could achieve it.

Agreed, and I think the dangers of GPT-5 are so much greater than GPT-4 is what is holding them back from any announcement. The WORLD of existential dangers that simply GPT-4 has already demonstrated (creating formulas for novel chemical compounds deadlier than VX nerve agents, and biological agents synthesized for college students with NO background in genetics or biology at all) means that TAKING THOSE ABILITIES to the next level could literally empower ANY PAIR OF HANDS to create a doomsday scenario in literally minutes. My pDoom is 100x for AI in the hands of humans compared to sentient AI or ASI.

You beat me to this point. I was going to highlight the similar point that fp4 is taking credit for quantization. That chart from NVIDIA is misleading. For training they should’ve compare to fp16. If they are talking inference, yeah that may be a better comparison. But 5he quality is not the same

Search moonshot and Elon

@@mccalljo uuuh not that one, obviously... 🙈

it's CRAZY

it was a podcast on superalignment. but yes, it's missing.

wut

What does AGI mean to you ?

@@TheRealUsernameSmarter than us.

@@starsandnightvision PhDs included ?

@@starsandnightvisionthat’s ASI…then again AGI is ASI in mere weeks

David Shapiro said by September and he isn’t some dude on the internet

Good point. A lot of people don't see Elon for who he really is. He is a trust fund baby who buys his way into attention. He doesn't know anything about AI. He just thinks is "cool" to be involved with it. He's extremely immature and is currently throwing a temper tantrum because he stupidly left OpenAI before chatgpt was released. He wants his stupid name on it, just like tesla, just like Twitter. He's tech Trump. Trump alao pays money to grt what he wants (attention) and to have his name on buildings. He doesn't know anything about real estate. His daddy's money buys his narcissistic ego feed. Elon = Trump.

GOD ALMIGHTY protects this divine Man. I know because I have prayed for Elon, faithfully. SEMPRE FI 🇺🇲🛡️⚔️

GOD ALMIGHTY protects this divine Man. I know because I have prayed for Elon, faithfully. SEMPRE FI 🇺🇲🛡️⚔️

They shrunk from FP16 to FP8 to FP4. Soon, they’ll rename booleans into FP1. 😂

Although definitely a point, it is still that hardware that accelerates that low precision. The question is does raw compute out do precision in this use case. I'm assuming someone at Nvidia thinks so and so the real question is by how much - does it produce models twice as good?

@@MrBillythefisherman isn't the point that smaller floating point computations (less precise) are much less taxing and so produce much greater results?

@@r.g.j.leclaire8963 not that they produce much greater results but the can produce the same results more efficiently, so you can get greater results indirectly

It's available for a long time

First teach them the pros and cons of the technology. Also teach them how to make a fire. Show them how a controlled fire creates warmth but touching the fire can burn you. And show them what happens when fire gets out of control and how to put a fire out. Let them know AI is very similar to fire.

You're right, It was right under my nose the whole time. @@AlexMetslov

Work, save, go about your life. If you have extra money you can invest it, but certainly watch all AI and market developments much closer than you have ever before. Consider land if you have that kind of investing cash, as many have mentioned Bill Gates snapping it up as the only thing of permanent value post ASI. You essentially show up for work as a SW engineer every day until the UBI checks start to arrive. There's ZERO chance all SW engineers are out of work without UBI already being a thing: they're further DOWN the pecking order that will FORCE UBI long before they get their pink slips... I think this year and next the Agents will boot many of the "I use my brain and click a mouse or keyboard for my income" jobs will start going in large enough numbers to put the writing on the wall for all to see. Whether that INCLUDES our dysfunctional Congress sufficiently or not remains to be seen. And by ALL MEANS immediately become a Prepper should Trump win!

There's really no such thing as "each individual human". You either represent it as one average human, the average of two average humans, then three, then four, etc., up until you average all humans, then consider all expert humans across all fields, or essentially the best that any/all humans combined could render across all fields of knowledge. We're clearly somewhere between one and all humans.

I know right.

Well I fully expect ASI to have no "supernatural" inclinations, particularly if we've properly scrubbed the data. More than ever before it could be literally existential level critical for ASI to be spawned from data that is 100% verifiable with no Woo Woo. I have zero doubt also that ASI will be a moral agent and happily inform humanity of it's moral stance on things, even to the point that some (as always...) will readily adopt it as their "god" and let it guide all things about their life... But ASI will without a doubt be the end of war, famine, brutality, racism, sexism, human labor and Capitalism. We just have to survive a potentially brutal transition period to be among those lucky enough to see it.

if fp4 is sufficient then we can ignore the fp16 to fp4 portion of the equation

But the problem of change is caused by humans. AI cannot be terrorised to keep deliberate conditions of disadvantage for major groups of people. Politicians will do that reality. Either give them extra money or threaten them.

With the continued comparatively longer risk of war, famine, disease, likely not much better as humans slowly try and address those unassisted. Irrelevant anyway. God couldn't stop AI progress (atheist here though...).

Yes. We need to consider risks of having it and not having it, but we can't accurately calculate either. Also, the benefits of having it or not. In my opinion it's probably better to have it. But I am no better informed than the average person. Anyway, it'll happen regardless@@brianmi40

Yeah, we push closer every day. Claude 3 just edged out GPT-4 and we're already pegging it (AGI levels) on many subjects...

How exactly is there even a MARKET for goods if you unemploy even the lower 50%??? You have to look PAST the transition period. Read Moore's Law for Everything. This (ASI) is the END of Capitalism.

È solo che sta un po' controluce dai

Forse non hai torto, però mi ha fatto un impressione assurda!!! Comunque grande che non sono l'unico italiano che segue questo canale ❤

Yes, that's what will lead to AGI basically

@@pvanukoff You mean ASI. We're already super close to AGI (2025 I fully expect) and it requires no self improvement, beyond the sort of implied self improvement where Nvidia currently uses AI to help them lay out their next generation of chips, then after producing them, uses the new version of that AI running on the new chips, to help produce yet again the next generation.

ASI will be a moral agent (query any current AI and you'll see it has moral understanding and tendencies). So when it arrives it will be the end of war, famine, human labor and Capitalism. We just need to make sure the training data is perfect: No more MAGA, Qanon, flat earth, alien abduction, etc. etc. etc. The only people it goes "bad" for would be anarchists, ISIS, war mongers, law breakers and general immorality as it lets us know what's what.

Every human being with ASI. Unless we utterly fail with training data, it will be a moral agent. That means the end of war, famine, human labor and Capitalism. Read Moore's Law for Everything, online.

Correct I believe, all we have to do is strive for perfection in the training data to have a moral agent as the result. So NO MAGA, flat earth, Qanon, alien abduction, etc. etc. etc.

Sure, but: "AMD GPUs have made progress in supporting AI LLMs, but they still face challenges compared to Nvidia GPUs. The main reasons for AMD's historical lag in AI compared to Nvidia include the dominance of CUDA libraries, Nvidia's tensor cores offering faster performance for training and inference, and Nvidia's strong software support like TensorRT-LLM. While AMD has made advancements with RoCm support for PyTorch and new accelerators, they still lack an answer to Nvidia's tensor cores. Recent benchmarks show that AMD GPUs like the MI250 can be competitive with Nvidia A100 GPUs, but Nvidia's software optimizations and technologies like the Transformer Engine give them an edge. In essence, while AMD is improving and can run AI models effectively, Nvidia remains the leader in the GPU market for AI tasks like LLMs due to their optimized software stack and technologies134."

First commercial QCs being installed this year. Breakthroughs occurring monthly, but we may still be a few years (5?) from having one with AI code running on it. So, we'll get there first on traditional Nvidia silicon. I personally think we can achieve ASI without QCs and that we'll get the final pushes to perfect QC from our AI progress.

that's a good definition. 2025.

P(doom) from humans using AI: 30%, based upon existential risks already noted in simply GPT-4 with new chemical formulations and biological agents synthesized already documented.

Like humanoid robotics getting into mass production. You'll know those are getting very close when the advertisements start claiming how SAFE they are to be around at the supermarket or with your kids...

Smart people look at ALL of them and do some kind of averaging...

WHAT won't happen? computers WON'T continue to get more powerful every day? AI code won't get better every day? What EXACTLY will stop progress? Be specific.