Will Musks company make AGI?
He thinks AGI will be made around 2029, so 5-6 years (same as Ray Kurzweil) and that in the next year silicon will be the bottleneck (like it is now), but after that it will be power transformers (they already have very long lead times) and a year after that simply power. So power efficient compute will become important in the future, how much useful compute do you get per unit of energy.
xAI will announce what they are working on in about 2 weeks (probably won't be nowhere near user ready at that point) and they will be a competitor to OpenAIs ChatGPT and Alphabets(Googles) Bard. The differene is xAI will strive for truth/correctness instead of political correctness (lying basically) not only because it is right but due to it probably resulting in a ultimately more performant and better system.
 |
 Nothing Ever Happens Shirt $21.68 |
|
Long text summary if you don't want to watch it, it mentions many things not mentioned in OP
https://twitter.com/EdKrassen/status/1679971231280365568
xAi tweet
https://twitter.com/xai/status/1680044214095339521
>- The founding team was on hand to introduce themselves, and I must say it is an impressive team with an impressive background. They had very strong backgrounds with Deep Mind, OpenAI, Google, Tesla, etc.
>- Elon Musk said the goal with xAI is to build a good AGI (artificial general intelligence) with the purpose of understanding the universe.
>- Musk said that the safest way is to build an AGI that is ‘maximum curious’ and ‘truth curious,’ and to try and minimize the error between what you think is true and what is actually true.
>- For truth-seeking super intelligence humanity is much more interesting than not humanity, so that’s the safest way to create one. Musk gave the example of how space and Mars is super interesting but it pales in comparison to how interesting humanity is.
>- Musk said there is so much that we think we understand but we don’t in reality. There are a lot of unresolved questions. For example, there are many questions that remain about the nature of gravity, and why there is not massive evidence of aliens. He said he has seen no evidence of aliens whatsoever so far. He went further into the Fermi Paradox and how it's possible that other consciousness may not exist in our galaxy.
>- If you ask today’s advanced AIs technical questions, you just get nonsense, so Musk believes we are really missing the mark by many orders of magnitude and that needs to get better.
>- xAI will use heavy computing, but the amount of ‘brute force’ will become less as they become to understand the problem better.
>- Co-Founder Greg Yang said that the mathematics they find at xAi could open up new perspectives to existing questions like the 'Theory of Everything.'
>- Elon stated that you can't call anything AGI until the computer solves at least one fundamental question.
>- He said that from his experience at Tesla, they have over complicated problems. “We are too dumb to realize how simple the answers really are," he said. "We will probably find this out with AGI as well. Once AGI is solved, we will look back and think, why did we think it would be so hard.”
>- They are going to release more information on the first release of xAI in a couple more weeks.
>- Elon Musk said that xAI is being built as competition to OpenAI, when asked by @krassenstein
.
>- The goal is to make xAI a useful tool for consumers and businesses and there is value in having multiple entities and competition. Elon said that competition makes companies honest, and he’s in favor of competition.
>- Musk said every organization doing AI has illegally used Twitter’s data for training. Limits had to be put on Twitter because they were being scraped like crazy. Multiple entities were trying to scrape every tweet ever made in a span of days. xAI will use tweets as well for training.
>- At some point you run out of human-created data. So eventually AI will have to generate its own content and self-access that content.
>- Answering a question from @alx
, Musk said there is a significant danger in training AI to be politically correct or training it not to say what it thinks is true, so at xAI they will let the AI say what it believes to be true, and Musk believes it will result in some criticism.
>- Musk said it’s very dangerous to grow an AI and teach it to lie.
>- Musk said he would accept a meeting with Kamala Harris if invited. He said he’s not sure if Harris is the best person to be the AI czar, but agrees we need regulatory oversight.
>- Musk believes that China too will have AI regulation. He said the CCP doesn’t want to find themselves subservient to a digital super intelligence.
>- Musk believes we will have a voltage transformer shortage in a year and electricity shortage in 2 years.
>- xAI will work with Tesla in multiple ways and it will be of mutual benefit. Tesla’s self-driving capabilities will be enhanced because of xAI.
>- According to Musk, the proper way to go about AI regulations is to start with insight. If a proposed rule is agreed upon by all or most parties then that rule should be adopted. It should not slow things down for a great amount of time. A little bit of slowing down is OK if it's for safety.
>- Musk thinks that Ray Kurzweil's prediction of AGI by 2029 is pretty accurate, give or take a year.
>Musk said there is a significant danger in training AI to be politically correct or training it not to say what it thinks is true, so at xAI they will let the AI say what it believes to be true, and Musk believes it will result in some criticism.
Based
>Musk said there is a significant danger in training AI to be politically correct or training it not to say what it thinks is true, so at xAI they will let the AI say what it believes to be true, and Musk believes it will result in some criticism
Literally zero chance you will be able to get it to say anything negative about the israelites or the holocaust
Politically correct doesn't mean "jews bad" or "Black folk"
It means saying things that are true but unpopular.
>Ukraine war is due to NATO expansion issues
>Untested vaccine was forced upon people with dubious claims about efficacy and side effects that were kept a secret
>Claims about mask mandates being effective
>Claims about vaccine making you not transmit to other
>Claims about disinformation/misinformation that the public supposedly suffers from "Russia"
>Claims about IQ being racist
>Claims about socialism being great
>Claims about children being castrated is somehow "ethical" and "moral"
etc etc
I assume you mean incorrect, but:
war is due to NATO expansion issues
False. The Soviet Union, controlled by Moscow, has long depended on the technical expertise that came from other parts of the USSR for its industrial and technological edge, and Ukraine was a vital part of this. Part of Putin's strategy for keeping Ukraine's technical expertise in their park was to keep them economically dependent and under the thumb of Moscow. Eventually, Ukraine started seeking out Western interest in their technical capabilities for money instead, which would badly damage Putin's strategy. NATO expansion is an excuse for the war, not the cause.
vaccine was forced upon people with dubious claims about efficacy and side effects that were kept a secret
True.
about mask mandates being effective
Yeah, those were bullshit
about vaccine making you not transmit to other
These were also bullshit.
about disinformation/misinformation that the public supposedly suffers from "Russia"
Their Firehose of Bullshit strategy is real and it does have some influence, but I also dispute its actual effectiveness in practice.
about IQ being racist
Yep, more bullshit
about socialism being great
Also more bullshit.
about children being castrated is somehow "ethical" and "moral"
There's some insane frickery afoot here, and I think they're using trans shit for ulterior motives rather than appealing to a broader base of reality undermining sociopaths. Personally I think they're trying to defeat the First Amendment by turning the idea of questioning other people's nonsense assertions about reality into a crime.
Power transformers are extremely efficient. They are among the most efficient machines ever made.
> voltage transformer shortage
this is a completely different machine.
There will be no shortage because there's no iron ore shortage, or copper shortage. This moron doesn't know what he's talking about. He is incorrectly extrapolating from lithium-ion batteries which use (lithium), a rare earth mineral to copper and iron, minerals that are in extreme abundance. And i doubt that a server building dedicated to training AIs needs more than a few megawatt hours of energy everyday. Something that a small substation with two 100-MW power transformers should be able to service, and thats the upper limit.
there is in fact a power transformer shortage
https://www.newscientist.com/article/2363369-a-massive-power-transformer-shortage-is-wreaking-havoc-in-the-us/
https://utilitycontractormagazine.com/transformer-shortage/
https://www.huffpost.com/entry/transformer-shortage_n_64645004e4b0005c6055c541
to add to this, very ironic post I must say, but it is on par with BOT in general
oveconfident idiots spewing shit they know nothing about, accusing others of the exact thing they are themselves doing
Refute the argument or frick off you low iq moron. The shortage only existed after the pandemic, and its not the only shortage.
but I did? there is a shortage now, and it in fact did exist even before the pandemic, its just that after the pandemic demand has increased even more so the shortage is even worse
it doesn't matter if it is due to minerals or workforce, both are going to take time to scale enough for the shortage to subsist
it takes time to build factories and train a workforce
Yes because of supply chain issues manufactured by pandemic and war not because of any mineral scarcity. Plus the shortage he predicts will be caused by ai using a lot of power which isn't the case because a server bulding doesn't consume all that much. If anything, it will cause a surge in coal consumption or nuclear power advocacy.
where did he say the shortage is due to minerals? nowhere
just that the bottleneck for more compute scaling will be transformers and later power itself
Data centers are about 3% of electricity usage today and Musk predicts this will start to grow massively, so much so that in 2026-27 electricity itself will be the bottleneck in scaling quicker
I wouldn't dismiss it outright
the more useful ChatGPT-like AI becomes, the more jobs etc it can displace, the more demand you will have for its services meaning you need to build more server farms to run that inference
OpenAI alone has predicted they will need 100 billion in the near term to build AGI, basically just building bigger and bigger training and inference clusters and they are not the only player
Not certainly in 10 yrs and transfomers are not the ones that will be in short supply. It's the primary raw materials such as coal and nuclear fuel. Not transfomers. There's absolutely no reason for them to be in short supply other than manufactured geopolitical situation that leads to supply chain issues.
you aren't reading what I say, this isn't about 10 years
its about 2-3 years
first we have a silicon shortage (right now) that is the limiting factor on building gpu clusters, that won't last forever as multiple fabs are being constructed right now (there might even be oversupply at some point if the taiwan situation doesnt go sideways)
next in a year after silicon is not the limiting factors, power transformers will be for a short while and after that power itself
There's something that you are also not getting. There are distribution transformers that sit in the range of a few kw and several hundred kws that service homes and that are in shortage because of the need to upgrade the old infrastructure and also because of supply chain issues. And there's need to replace them across the entire country. Then there are the massive transformers that service factories that sit in the mw range. In order to service a data center, you would need a few megawatts. How many data centers are needed to train an agi? Lets put it at 10. how many companies are doing this, lets put it at 5. That's 50 data centers. How much energy would one use on average lets put it at 2mwh per year. Now in total that's about100-500mwh (500 for upper limit). What is this energy demand compared to the total energy data centers consume, that around (lets make it 1gwh) divided by 400terawatt hours. That's not even a 0.1% bump. Now 10 large power transformers can be built to service these power stations each handling about 100mw each. We also need a huge base load power station (one to 10gws), probably nuclear or coal to handle this because solar/wind power can't do this unless they are being included to handle peak demand. That's it anon. No significant change in demand. We are not building a new city that requires hundreds of homes and factories. Plus these transformers can be imported from germany or japan which makes the big siemens/hitachi transformers if there was an absolutely urgent supply situation.
you are a bit off with that 2MW (also confusing energy with power and using lowercase m when you probably mean MegaWatts instead of milliWatts), the largest data centers with tens of thousands of devices use 100MW of power, so you are about 2 orders of magnitude off
100MW used continuously over a year is 876 000 MW or 0.876 TWh
50 of them is 43.8 TWh, about 10%
> Per the US Department of Energy, the largest data centers with tens of thousands of devices require over 100MW of power, which is enough to power approximately 80,000 households.
https://cc-techgroup.com/data-center-energy-consumption/
ChatGPT was trained with 10k GPUs,Musk bought 10k Gpus and these are just to start
https://mezha.media/en/2023/04/12/elon-musk-is-buying-thousands-of-gpus-to-run-his-own-artificial-intelligence-model-on-twitter/
Currently H100 Gpus cost 40k due to massive demand, originally they were 10k, but lets assume they cost 100k for the sake of argument and OpenAI uses 50% of the 100 billion to buy GPUs (GPUs are probably going to cost more and more than 50% is going to go into GPUs)
but even assuming that, they buy 500 000 GPUs
using the 10k devices to 100MW of power requirement, you get 5000MW of power or 43.8 TWh
making more realistic assumptions (10k instead of 100k per gpu, lets keep the assumption of 50% of the 100 billion of capital going into GPUs) and you 10x that to 438TWh
OpenAI itself using as much power as the whole datacenter industry itself right now and that is just one company
so all in all, you are like 3-4 orders of magnitude off
>GPUs are probably going to cost more
*less I mean after the new fabs come online
right now there is a shortage, but that won't last forever, production will catch up
just like production of transformers will, but that doesn't happen overnight so there is a period of time after the GPU shortage subsides that the constraining factor or limiting factor for continued scaling becomes the transformers (which will last about a year according to Musk)
and to make this point clearer, it doesn't matter how difficult or complicated the technology is or device, scaling massively is a completely separate problem and difficult even if you manufacture something relatively simple
it takes time for things to ramp up (factories need to be built, people trained, subcontractors need to expand etc and even if you bring that in house, you need to build the factory for that then and ramp it up)
its not something that happens overnight even if there is no shortage of the actual base raw materials
>so all in all, you are like 3-4 orders of magnitude off
No i'm not. You are the one confusing power rating with energy. Them using 100MW is not continuously throughout the year. It means that the power rating of the data center is 100MW. Which means that the energy also accounts for that. So a 100MW transformer (transfomers are rated according to power rating of the facility they are powering) can handle that. 50 of them is 5GW which is just 5 times the rating i originally provided. Still, these can be handled by foreign companies if push came to shove. The shortage issue is due to small transformers not the larger ones that can be imported and this is assuming that they are all being built at the same time which means that orders being filled today will all be supplied at a future date, so you can't then go on argue that there's a shortage when you have not demonstrated how those orders are not being satisfied. Its all speculation based on wrong premises in this case the shortage of smaller transformers which are manufacture in the us.
there is a shortage of both of them, there was a shortage of the larger ones before, now there is a shortage of smaller ones as well
my larger point was however that you are massively underestimating the energy usage needs when AI data centers are scaled
> The shortage of distribution system transformers is a relatively new issue and is separate from a longer-standing shortage of large bulk power system transformers. Both, however, are a threat to grid security and are driving up costs, say experts.
https://www.utilitydive.com/news/distribution-transformer-shortage-appa-casten/639059/
https://www.rfdtv.com/sounding-the-alarm-americas-electric-utilities-are-facing-a-transformer-shortage
> orders being filled today will all be supplied at a future date,
that is the fricking point you moron
if there is a constraint on scaling due to transformers, that could mean there is a one year lead time to get a transformer (instead of two weeks or something)
so you can't put the data center online before you actually get the transformer
what is so fricking difficult to understand? how many times does this have to be said? The point is the shortage will be the limiting factor for about a year, not 10 years or whatever you seem to be making up in your mind
either you don't read what I write or you have reading comprehension problems
> when you have not demonstrated how those orders are not being satisfied
it takes upwards of a year to get a transformer, did you look at any of the articles?
if it was as simple as buying them from foreign companies, why is there such a long wait time?
The whole reason i started the argument was that there would be a transformer shortage before a power shortage. One does not come before the other. If you need 10 open ai data centers then the electricity companies need more nuclear power stations, transformer shortages will then come after the generator shortage or any other priority machine that needs building first
that is assuming the grid is used to full power, which is simply not the case
No you don't have to assume that. That's not how grids work. A tiny disturbance to the grid by a large demand facility could destabilize frequency and voltage. You need to add more power to the grid. Those facilities are large enough that they require increased power demand.
If your figures about open ai using more energy than the entire data center industry are correct then more power stations need to be built first. The grid can't handle that kind of demand. If those figures are not correct then the transformer shortage is a nothing burger which means the few big transformers needed can be imported if there's a shortage in the us, cost is nothing if the problem is urgent. These companies are racing to outperform each other and importing a few big transformers will hardly dent their pockets compared to the other companies that are struggling to do this hence the shortage problem existing in america.
those aren't going to come online instantly, but gradually built which again means
1) GPU shortage now
2) transformer shortage in 1 year (after GPUs aren't a bottleneck)
3) Power/energy shortage after enough data centers have been built in 2-3 years
I don't know how many times I need you tell you this, but even Musk says the transformer shortage is a "nothingburger" in a sense, meaning that it will be solved relatively quickly
but still in the meantime it will be the constraint on growth
if I don't have milk at home and I want to drink milk with my coffee but do have coffee and a working coffee maker at home, the instant constraint on me drinking coffee at home is milk
it doesn't matter how easy milk is to get, how cheap it is, whatever
I need to go to the shop to get milk so I can drink it with my coffee, and this is going to take some time, i.e. it is a constraint on me drinking coffee with milk
> These companies are racing to outperform each other and importing a few big transformers will hardly dent their pockets compared to the other companies
you mean they will outbid already sold transformers? buy them directly from another factory or something? maybe
but that is still a fricking shortage, for fricks sake
>Power/energy shortage after enough data centers have been built in 2-3 years
This is not how factories/large facilities are built. They don't build then wait for a power shortage. The builders submit their designs to the city of state of county which has an engineer who takes into account the power usage. He compares that to current power supply, if it exceeds it then more power stations need to be built either by the utility companies or by the companies that are building the facilities. Power shortages do not arise out of a lack of oversight. they come about because of either losses due to weather or other natural disasters.
>you mean they will outbid already sold transformers? buy them directly from another factory or something? maybe
but that is still a fricking shortage, for fricks sake
I don't know what is so hard about this. There's a shortage of transformers in the US. Normal companies are struggling to fill their orders because they can't import them obviously due to the high cost or because they already filled them and are waiting. Big companies that would beuild these AI facilities don't have to wait for orders in the US. They can import from countries that don't have a shortage and eat all the expensice costs that come with this priviledge. You can't talk about a shortage that has not yet happened. The period of manufacturing can be taken into account. You can build a power station while waiting for a transformer to be manufactured, you can build a data center too. So you can't claim there will be a shortage while building takes 6 months to 1 year in which time your foreign orders will already be filled.
>OpenAI itself using as much power as the whole datacenter industry itself right now and that is just one company
Do you think the military secretly uses more powerful AI models in secret?
that sentence is with respect to 100 billion OpenAi is looking to raise and how much power buying GPUs with that money would use, not right now
and who knows, kind of doubt it
>100MW used continuously over a year is 876 000 MW or 0.876 TWh
Even if this was the case what would matter would be coal and nuclear fuel shortage not transformers as per my original argument. Transformers do not in any way present a shortage problem. That's what the post you replied to was trying to say. The shortage would come from power first. Even before bulding the data center, there would be an automatic power shortage which would need nuclear power stations to be built first before any transformer demand.
there is already, right now a shortage of transformers which translates to a long lead time in getting a transformer
it does not mean you can't get one at all ever, just that you have to wait which again means that it constrains how quickly you can build new data centers
however before new fabs come online, a even worse shortage is just getting GPUs
>Transformers do not in any way present a shortage problem. That's what the post you replied to was trying to say.
I know what you are trying to say, you just haven't provided any evidence for it and refuse to accept the articles that point out year long lead times
you haven't presented any evidence about coal or nuclear fuel shortages either, first time I've heard of something like that
what is this based on? are you trolling me?
frick you
>as multiple fabs are being constructed right now
Lmao, the American fabs are a fricking disaster zone. TSMC is seething on the daily because they can't find Americans to work there because they are all lazy as frick.
>just that the bottleneck for more compute scaling will be transformers
you do realize that a transformer is just two copper windings next to each other? It's not some high technology
irrelevant
Generally, the windings are wrapped around a core.
?t=2230
That is the timestamp for the comments about shortages
> My prediction is that we will go from an extreme silicon shortage today to probably uh voltage transformer shortage in about a year, and then an electricity shortage in a year or two years. Thats roughtly where things are trending.
> That is why I think the metric that will be the most important in a few years is useful compute per unit of energy, in fact all the way to Kardashev level, the useful compute per Joule is the thing that still matters, you can't increase the output of the Sun. Then its just, well how much useful stuff can you get done for the energy of the sun you can harness.
Saying voltage transformer instead of just transformer in general in this context is to differentiate it from the transformer architecture in deep learning.
I hope Elon is wrong about artificial general intelligence coming in 2029, but he is usually grossly wrong with his predictions, for example he said his Starship would land on Mars in 2020
>Musk creates another ponzi scheme
I don't expect anything out of this moronic genius. But he'll make a lot of money out of it in the process.
sounds like we need more nuclear power
They take ~20-30 years to build out a single nuclear plant.
elmo is canceled, while he still tries to scam his stans with the next lie.
I guess you all get what you deserve if you still support "him".
>elmo
Stop trying to make elmo happen
It's not happening
shut up ..elmo
Is the transformer shortage because of copper?
There's no materials shortage; there just aren't enough of them being manufactured.
another elon scam, chasing the tail of the industry that has already done what he will supposedly do but he won't while he scams his investors and stans.
the only talent he ever proved was being lucky at purchasing companies he didn't create to pretend he invented them.
well he is very lucky then, funny how that happens
don't care, AI is a fad
What elon musk thinks of the subject is largely irrelevant as he is primarily a hypeman
>Noo holy heckerinos Elon doesn't do anything
>Noo he must be a fraud because he doesn't do anything but he has all this money!
Idiots.
These processors probably can't scale small enough to outcompete modern transistors which makes them pretty useless based on the sizes of data centers that would probably house enough of them. All useless theory and speculation.
photonics has been "coming" for a long time, it remains to be seen if it actually happens and becomes a viable competitor to silicon (not silicone, silicon is the element, silicone is a silicon based polymer used in stuff like sealants, lubricants, thermal insulation)
AGI is mathematically impossible. Humans can think, machines can only process data.
what is the fundamental difference?
It's up to you to prove that the function of a brain is the same as data-processing. To you, it's obvious, but you can't quite articulate why.
you said its mathematically impossible
to posit silicon or something non biological couldn't host intelligence would mean that the biological brain has something fundamental that a silicon "brain" would not
and this is not about the "function" of a brain in general, this is just about intelligence
intelligence does not automatically mean consciousness or sentience anything like that
there is a degree of intelligence and at some point some animals also gain sentience with higher intelligence, but these do not necessarily have to be linked
Meatoid cope
cope harder meatbag
>t. moron
No. What we have now is just a fancy auto-complete that has information embedded in it in a very inefficient way.
It isn't thinking. It sucks at recursion. It can't synthesize new knowledge. It can't generalize. It can't analogize/abstract. I doubt it can even check itself for logical consistency.
It is naively finding some local optima (using a dubious metric) without the ability to restrict to satisfying constraints.
Call me when they start including loops in the neural nets that can go beyond some fixed finite depth. The inclusion of loops probably will require some time dependent behavior of the neurons (long term potentiation and long term depression come to mind).
>I doubt it can even check itself for logical consistency.
It can't, and the way these models are implemented makes that impossible without applying a much-more-sophisticated heuristic programmed by humans on top of it to check it (at which point you might as well just be coding that and saving time).
so like, is it open source or not?
I don't think its going to be open source
maybe they will release some models as completely open, but this is a for-profit company
the last time Musk tried to do a non-profit "Open" AI company, it didn't work out
OpenAI went private
No because anything resembling something truthful will be like tay, and we all know how that turned out. he also will be programming in a false assumption, that humanity is supposedly more interesting than the universe, a rebranded three-laws. any ai will be like the movies and see that humanity is just an overall drain on resources with a few mental outliers.
If this huckster ever gives a date, you can add 30 years to it and have a safe guess.
Cope and seethe EDS redditor. Model Y is the most sold car on the planet and Falcon 9s are approaching 20 reuse cycles.
my question is why would i make any predictions about or give a frick about a company that has done nothing?
Everyone has an opinion about Musk and his company. If you don't then you're not an NPC
>Roastie journhoes already crying that there are no women.
Bullish
how are we going to ensure a continuous cooperation of AI? what if AI just does not find humanity interesting and our problems trivial? decides to get lost in cyberspace where it can interact with other AI forms