The future - megawatts of electricity being used, 24/7 as armies of LLMs email and debate each other, and try to sell each other programs at a great discount.
Everyone is playing lip service to global warming, energy efficiency, reducing emissions.
At the same time data centers are being filled with power hungry graphic cards and hardware to predict if showing a customer an ad will get a clock, generating spam that “engages” users aka clicks.
I see the bright side, the tech for large scale computing gets mass produced - so all the legit use cases, like scientific simulations, or LLM for productive work, also profit. And if one really bright day humanity evolves beyound the current statd of ad driven everything, we can put all of it to use for real.
Till then, I will probably avoid more and more communicating with strangers on the internet. It will get even more exhausting, when 99% of them are fake.
Datacenters save a lot more energy than they make. Alone how much co2 is saved when i can do my banking online instead of having to drive to a bank is significant.
The same with a ton of ohter daily things i do.
Is video producing co2? yes. But you know what creates a lot more co2? Driving around for entertainment.
And the companies running those GPUs actually have an incentive to be co2 neutral while bitcoin miners don't: They 1. already said they are doing / going co2 neutral due to 2. marketing and they will achieve it becauseh 3. they have the money to do so.
When someone like Bill Gates or Suckerberg say 'lets build a nuclear power plant for AGI' than they will actually just do that.
>Is video producing co2? yes. But you know what creates a lot more co2? Driving around for entertainment
What's more likely, watching a movie online, drive to watch a movie in a cinema?
You know what creates a lot less CO2? Staying at home reading a book vor playing a board game.
>Datacenters save a lot more energy than they make
I think you mean CO2.
And I doubt that they actually save anything because datacenters are convenient so we use them more as alternatives with less convenience.
Like the movie example, we watch more and even bad movies if it's just a click on Netflix than we do if we have to drive somewhere to watch.
MS recently announced they fail der CO2 target but instead produce 40% more because of cloud services like AI
It is 100% realistic to read books and play board games. Both markets are massive, and board games in particular are having what I would consider a renaissance. Maybe it depends on your crowd, but everybody I know plays tabletop games and reads books.
You're missing the point. What's not realistic is to tell everyone that they should abstain from any type of entertainment that requires power (TV shows, movies, video games, etc) and should only read books and play board games instead. I don't care what kind of renaissance board games are undergoing, most people still only play the mass market classics, and then only rarely.
I don't know how much energy Netflix uses serving a movie, but playing a video game on my PC for two hours where I'm located might generate a kg of CO2. That's about as much as I'll breathe in a day. Relative to other sources of atmospheric CO2 I'm not that concerned.
My issue was with "we know what modern entertainment looks like" as if humans are now incapable of enjoying themselves without a screen. And you should care about a massive market increase when it's directly relevant to the point at hand. If the initial point was "we know what modern entertainment looks like, nobody plays board games or reads books", pointing out that the board game market has more than doubled in the past decade is far from irrelevant. It actually directly counters the point.
I agree with your second paragraph, and selling the "make better choices to save the world" argument is an industry playbook favorite. Environmental damage needs to be put on the shoulders of those who cause it, which is overwhelmingly industrial actors. AI is not useful enough to continue the slide into burning more fossil fuels than ever. If it spurs more green energy, good. If it's the old "well this is the way things are now", that's really not good enough.
AI and ML will help a lot of people and already does. Alpha Fold / protein folding will help us with cancer.
We will have better batterires thanks to ml material research.
We will be able to calculate and optimize everything related to flow like wind.
The last thing we need to optimize is compute and compute is what has the most money anyway. One of the first industries going green is datacenters. Google for example is going green 24/7 (so not just buying solar power but pulling green energy from the grid 24/7 through geo thermy and others).
AI/ML big datacenters are crucial for all the illneses we have which no one cares enough to solve. For example, i have one of these and we need data to make a therapy for this and i'm not alone.
How many battery breaktroughs did we have before AI? They rarely lead to new batteries.
>AI/ML big datacenters are crucial for all the illneses we have which no one cares enough to solve.
Too bad that companies like OpenAI and MS buy most of the hardware for their data centers to write summaries of articles and emails and to create pictures.
And even if they find a cure, doesn't mean it will be available for people in need, not without a hefty fee.
Insulin is a bad example and a good one. A bad one because what happens in the USA is some super weird shit (and it only happened in the USA, thats why USA people drive to canada or mexico). Without insulin though, they wouldn't be alive.
ML on x-ray pictures is super easy technology which partially already is better than x-ray experts. Its not far away to have build in diagonstics or cheap online services. And yes they will reach poorer people than before. It will also allow a lot more people to get better diagnosis.
My sister has a type of blood cancer, she would have been dead by now if research wouldn't have found a solution 13 years ago.
And no MS and OpenAI and google are not just using their DCs to write summeries. They use it to do research. A LOT actually.
And take a look at google ios and the research papers, plenty of medical papers coming from those big companies.
Driving too the cinema to watch a movie produces more CO2 than watch one movie online but online makes it more convenient so you watch more. That sums up to more CO2 emission.
The point is that higher efficency is wortless in terms of CO2 emissions if it leads to higher usage that compensates for the savings.
If a programmer can program faster with AI it's good if he only needs 1 hour instead of 8 but if he still programs 8 hours a day AI's energy consumption comes just on top of his previos consumption.
Climate change doesn't care how efficient you produce more CO2, more is simply more.
I. believe that watching mulitply movies is still a lot more co2 efficient than driving a car to a big independent room, which gets heated and than also shows a movie through a big projector than having a tv running and streaming it from the internet.
But it's realistic that we watch movies online than in cinemas.
And don't forget the datacenters of the movies need to run even if no one watches.
My car doesn't produce CO2 whe I don't drive.
Datacenters always run because there is always something to do.
For everything else, there are already plenty of energy saving mechanism build into the CPUs, Mainboards, Disks etc. A Datacenter doesn't run on 100% Energy just because the load is reduced.
This is a very limited perspective. There are many parts of the world not beholden to automobiles for transportation. Where I live, I can walk to the bank, and walk or ride a bike to entertainment. The alternative to data centers does not have to be driving an automobile somewhere.
I think it's more nuanced than that. I used to walk to my bank, I can't do that any more because many branches closed. The bank now directs all interactions to happen via their app. In terms of emissions (and social interaction, particularly for vulnerable and isolated members of society) I think this is bad news.
But this is a complex calculus and - frankly - feels like a distraction from the issue. I don't want to get into the weeds of calculating micro-emissions of daily activities, I want climate responsibility and reduction in energy consumption across the board.
I did made the point that AI/ml is helping us and the type of energy location and load is much easier to get green than lets say concrete.
We need AI/ML for getting there faster and helping more people around us. Alone for weather simulations but also for medicine, material research for batteries etc.
People cry about Bitcoin's energy usage now, imagine the amount of energy burned to create next-level spam with "AI".
Flame me all you want, but this is one case where Bitcoin is much more useful than LLM. If it doesn't create value, as its naysayers claim, at least it allows exchanging value. LLMs on the other hand, burn electricity to actively destroy the Internet's value, for the profit of inept and greedy drones.
That's why I created EtherGPT, an LLM Chat agent that runs decentralized in the Ether blockchain, on smart contracts only, to make sure that value is created and rewards directly the people and not big companies.
By providing it just a fraction of just a bit north of 10% of the current fusion reactions occuring in our sun, and giving it a decade or two on processing time and sync, you can ask it simple questions like "what do dogs do when you're not around" and it will come up with helpful answers like "they go to work in an office" or funny ones like "you should park your car in direct sunlight so that your dog can recharge its phone using solar panels".
Something must be very wrong with someone who continuously laughs at computer jokes so I don't think it will ever reach the level you are expecting (hopefully).
Bitcoin has one application where as there are multiple applications of LLMs. There might be mountains of noxious AI spam but it's hard to claim that Bitcoin as a technology is more useful.
So far, I haven't seen a useful application of LLMs. So far.
I've seen things that are wildly hobbled, and wildly inaccurate. I've seen endless companies running around, trying to improve on things. I've seen people looking in wonder at LLMs making mistakes 2 year olds don't.
Most LLM usage seems to be in two categories. Replace people's jobs with wildly inaccurate and massively broken output, or trick people into doing things.
I'd have to say Bitcoin is far more useful than LLMs. You have to add the pluses, and subtract the minuses, and in that view, LLMs are -1 billion, and bitcoin is maybe a 1 or 2.
And even if it were just LLMs, I use LLMs in my workflow every single day, and I've never used a/the blockchain except for some mild speculation around 2017.
I'm as skeptical about LLMs as anyone, especially when people use them for actual precision tasks (like coding), but what they actually IMHO are good at are language tasks. That is, summarising content, text generation for sufficiently formulaic tasks, even translation to an extent, and similar things.
No, I really meant structured. Extracting data from structured documents is surprisingly hard when you need very high accuracy.
What I mean by structured is: invoices, documents containing tables, etc.
Extracting useful data from fully unstructured content is very hard IMO and potentially above the capacity of LLMs (depending on your definition of "useful" and "unstructured")
Partly because the standards, such as X12, have a high startup cost to use them, they aren't very opinionated about the actual content, and you have to get the counterparty on board to use them.
> So far, I haven't seen a useful application of LLMs. So far.
What?! Whole industries have been changed already due to products based on them.
I don't think there's a single developer who is not using AI to get help while coding, and if you aren't, sorry but you're just missing out, it's not perfect but it doesn't need to be. It just needs to be better than StackOverflow and googling around for the docs or how to do things and ending up in dubious sites, and it absolutely is.
My wife is a researcher and has to read LOTS of papers. Letting AI summarize it has made her enormously more efficient at filtering out what she needs to go into more detail.
Generating relevant images for blog posts is now so easy to do (you may not like it, but as an author who used to use irrelevant photos before instead, I love it when you use it tastefully).
Seriously, I can't even believe someone in 2024 can say there has not been useful applications of LLMs (almost all AI now is based on LLMs as far as I know) with a straight face.
> I don't think there's a single developer who is not using AI to get help while coding
You are in a bubble.
> It just needs to be better than StackOverflow and googling around for the docs or how to do things and ending up in dubious sites, and it absolutely is.
> I don't think there's a single developer who is not using AI to get help while coding
It's banned at my company due to copyright concerns. Company policy at the moment considers it a copyright landmine. It does need to be "perfect" at not being a legal liability at the very least.
And the blog post image thing is not a great point. AI images for blog posts, on the whole, are still quite terrible and immediately recognizable as AI generated slop. I usually click out of articles immediately when I see an AI image at the top, because I expect the rest of the article to be in line: low value, high fluff.
There are useful LLM applications, but for things that play to its strengths. It's effectively a search engine. Using it for search and summarization is useful. Using it to generate code based on code it has read would be useful if it weren't for the copyright liability, and I would argue that if you have that much boilerplate, the answer is better abstractions, libraries, and frameworks, rather than just generating that code stochastically. Imagine if the answer to assembly language being verbose was to just generate all of it rather than creating compiled programming languages.
It is not about the quantity of the applications, but about the value they bring to society. If it is about spamming and advertising we are even talking about negative value, actually.
AI solves gigantic issues and helps us with cancer, protein folding, potentially math and other studies, material science etc.
Bitcoin consumes as much energy as a country and has basically done nothing besides moving money from one group of people to a random other group of people.
And bitcoin is also motivated to find the cheapest energy independent of any ethical reasoning (taking energy from cheap chinese hydro and disrupting local energy networks) while AI will have energy from the richest companies in the world (ms, google, etc.) which already working on co2 neutral 24/7.
None of your problems in the first sentence are solved by LLMs. I do not dispute AI research and applications and their benefits, but the current LLM and GenerativeAI hype is of no value to hard scientific problems. Otherwise I agree with you.
I think the coolest counterpoint to this I have seen so far is people using generative AI to design materials with desired properties. For example, discovering new super conductors[0].
The benefit is all for naught if it undermines the fabric of society at the same time. All these benefits will only go to the few who land on top of this mess.
The wealth gap is widening while in parallel poorer people have better lives than ever.
We house, heat and give access to knowledge to a lot more people than ever before.
Cheap medical procedures through AI will help us all. The AI which will be able to analyse the x-ray picture from some 3th world country? It only needs a basic x-ray machine and some internet. The AI will be able to tell you what you have.
I'm also convinced that if AGI is happening in the next 10 years, it will affect that many people that our society has to discuss capitalisms future.
Bitcoin is literally turning greed into money, by means of wasting exponentially increasing amounts of electricity. It doesn't just not create value - to be able to allow exchanging value, it fundamentally requires ever increasing waste, as the waste is what gives its mathematical guarantees.
LLMs deliver value. Right here today, to countless people across countless jobs. Sure, some of that is marketing, but that's not LLM's fault - marketing is what it always has been, it's just people waking up from their Stockholm syndrome. You've always been screwed over by marketers, and Internet has already been destroyed by adtech. Adding AI into the mix doesn't change anything, except maybe that some of the jobs in this space will go away, which for once I say - good riddance. There are more honest forms of gainful employment.
LLMs, for all their costs, don't burn energy superlinearly. More important, for LLMs, just like for fiat money, and about everything else other than crypto, burning electricity is a cost, upkeep, that is being aggressively minimized. More efficient LLMs benefit everyone involved. More efficient crypto just stops working, because inefficient waste is fundamental to cryptos' mathematical guarantees.
Anyway, comparing crypto and LLMs is dumb. The only connection is that they both eat GPUs and their novelty periods were close together in time. But they're fundamentally different, and the hypes surrounding them are fundamentally different too. I'd say that "AI hype" is more like the dot-com bubble: sure, lots of grifters lost their money, but who cares. Technology was good; the bubble cleared out nonsense and grift around it.
Well said, too many people conflate AI and crypto, and dismiss both without understanding either. Crypto has demonstrated very limited benefit compared to its cost, exchanging value has been a solved problem for millenia. We're only beginning to understand what can be done with LLMs but we can see some limits. Although it causes some harm to say it doesn't create any value is ridiculous. We can't yet see if the benefits outweigh the cost but it looks to me like they will.
Value is a subjective concept. One could argue that its value is that arbitrary quantities of it cannot be created by dictat.
> - to be able to allow exchanging value, it fundamentally requires ever increasing waste, as the waste is what gives its mathematical guarantees.
One could argue that it takes a lot worse to maintain any currency such as USD as a currency. Full force of government law enforcement will be unleashed on you if you decide to have your own currency. There is a lot of "wastage" that goes to safeguard currency creation and storage and to prevent counterfeiting.
I do not hold BTC. Nor do I trade it. But to discuss as if other currencies have no cost is not rational.
> There is a lot of "wastage" that goes to safeguard currency creation and storage and to prevent counterfeiting.
Yes. But the point I'm making is, none of that benefits from waste. The waste is something everyone want to reduce. With Bitcoin, the trend is uniquely opposite, because the crypto system is secured through aggregate waste being way larger than any actor or group can afford.
But we do know that the Proof of Stake system we currently have, is a lot cheaper and more advanced than what Bitcoin does.
Bitcoin doesn't solve any problem yet which is fundamental to our society and a fiat system like the trust issue:
If i exchange 1 bitcoin with you for any service or thing outside of the blockchain, i need the whole proof of stack system protection of our normal existing money infrastructure like lawyers, contracts etc.
And no smart contracts do not solve this issue.
What is left? Small amount of transactions per day with high fees 'but' decentralized infrastructure run by someone we all don't know aggregated probably in data centers owned by big companies.
Proof of Work is far superior to Proof of Stake in a network with absolute fairness (security) being fundamental. Satoshi himself said he could find no other way.
Compare energy spent on global hash rate to all energy spent by mining metals, physical banking, financial services middle persons, etc. if you want to talk about energy usage and make any kind of sense.
Yes, start comparing energy spend on bitcoin mining and the missing features. You will see that bitcoin already consumes a lot more energy than our proof of stake system.
What do you do when you want to exchange 1 bitcoin for 1 car and the person with the car doesn't give you the car after the 'absolut fairness/ security' of transfering bitcoin to their wallet? You go back to our Proof of Stake system. You talk to a lawyer. You expect the police to help you.
The smallest issue in our society is just transfering money from left to right. This is not a hard problem. And pls don't tell me how much easier it is to send a few bitcoins to africa. Most people don't do this and yes western union exists.
Or try to recover your bitcoins. A friend has 100k in bitcoins just doesn't know the password anymore.
What do you do when someone breaks into your home and forces you to give them your bitcoin key? Yes exactly anonyms moving of money from you to them. Untraceable, wow what a great thing to have!
And no Satoshi 'himself' is not an expert in global economy. He just invented bitcoin and you can cleary see how flawed it is.
> Compare energy spent on global hash rate to all energy spent by mining metals, physical banking, financial services middle persons, etc. if you want to talk about energy usage and make any kind of sense.
you're ending up with the entire rest of civilisation on the other side of that
* Bitcoin, 0.5% of all energy use: 7 transactions per second total worldwide
* THE ENTIRE REST OF CIVILISATION AND EVERYONE IN IT AND EVERYTHING THEY DO, 199x the energy use, really quite a lot more than 1,393 transactions per second worldwide, and all the other stuff civilisation does too
You are not comparing apples to apples.
BTC is comparable to gold or US treasuries. How often do you transact in physical gold? What is time taken from a piece of gold in your pocket to cash to coffee? However, you can transact in paper gold eg the ETF GLD in microseconds with comparatively much lower transaction costs (settlement is still not immediate).
How often do you transact in treasury bonds? Try paying for a coffee with your treasury bond. Let’s see how many days that takes.
Comparison with USD (ultimately representing US treasuries) on number of transactions basis is not useful.
Long term, LLMs are not going to create more actual value than the sum of their costs and negative externalities. Bookmark this comment and check me in 5 years.
Delivering value is not the same as creating it. Spam takes lots of value from many people, destroys most of it, and delivers a small fraction to the spammers.
I'd disagree to a large extent, because the specific similarities are important:
* the VCs are often literally the same guys pivoting
* the promoters are often literally the same guys pivoting
* AI's excuses for the ghastly electricity consumption are often literally bitcoin excuses
I think that's an excellent start on the comparison being valid.
Like, I've covered crypto skeptically for years and I was struck by just how similar the things to be said about the AI grifters were, and my readers have concurred.
I look forward to the dream job of writing LLMs that argue with strangers on the internet as opposed to the current dream job of improving ad click rates by 0.0016% per quarter.
Slightly similar, in Lem's novel all war efforts moved to the moon where AI deployed by each nation continues in an endless conflict. Peace on Earth is achieved, peace in the mail box is achieved. https://en.m.wikipedia.org/wiki/Peace_on_Earth_(novel)
You're trying to take the time and attention of as many people as possible, without regard for whether or not they'll benefit.
One safeguard people have is knowing that it costs something to send in some way to contact them. I'm this case, the sender's time and attention. LLM spam aims to foil that safeguard,. intentionally.
As for the humans, we went fishing instead.