But the spend on AI is globally is still measured in the tens of billions? Tiny in the grand scheme of things. So what 'money' is moving up? Not revenue, and in the case of a bubble bursting, not speculative capital.
Speed is not a problem for me. I feel they are at the right speed now where I am able to see what it is doing in real time and check it's on the right track.
Honestly if it were any faster I would want a feature to slow it down, as I often intervene if it's going in the wrong direction.
One of the biggest factors for me personally was going vegetarian, and then vegan. I didn't realize it for 30 years, but it's hard to feel connected to nature, animals, and the environment when you are eating something you didn't kill yourself. Once I made that move, it's a beautiful feeling and a kind of connection to animals and the planet I never knew before. I wasn't even much of a pet person before that.
I have no problem with the concept of lab grown meat, or 'cruelty free' meat as it's sometimes known. However, unfortunately it isn't yet completely cruelty free as it uses fetal bovine serum (FBS) to kick start the process.
Honestly perhaps you should have tried going fully plant-based. I truly think dairy is the worst of both worlds. It's too easy to lean on cheese which is not nutritionally balanced. Dairy can also inhibit iron intake which might be why you felt tired.
The good thing about a plant based diet is that it forces you to eat a ton of vegetables, fruits, nuts, seeds etc. I try and follow the Tim Spector recommendation to get 30 different plants and vegetables per week and easily exceed that these days. I never felt better, and my blood vitals are fine.
(yes I take a B12 supplment as all vegans must, but B12 is artificially added to cattle feed anyway as most modern cows don't make enough)
All that said, I also just don't think it's natural to consume breast milk from another animal. There are literally legal limits to how much pus (somatic cell count) can be in each liter/gallon of milk. No thanks.
That's my experience as well. I think vegetarianism is too much of an ideal and ideals can be very dangerous.
Also, we can't be sure yet since nutrition "science" is almost indistinguishable from quackery but I have the strong feeling that eating only plants has some dysgenic effect. When you look at nature, the smartest animals are predators because it allows them to "delegate" the hard work of converting low quality food to something packed with energy and other necessary nutrients.
Then there is the issue of food security and virtuous cycle, much of large agriculture depends on fossil fuels and it's questionnable if we should let go of animals provided inputs.
Perfect is the enemy of good as we say and vegetarianism/veganism might be one of those cases I think.
> Then there is the issue of food security and virtuous cycle, much of large agriculture depends on fossil fuels and it's questionnable if we should let go of animals provided inputs.
My understanding is that the production of animal products is far more environmentally damaging and land intensive than plant agriculture. Don't forget that a large amount of plant agriculture actually exists to produce feed for animals, so a reduction in producing animal products has an accompanying reduction in plant agriculture. There would need to be an increase in plant agriculture to replace the calories/nutrition not coming from animals (although addressing food waste would mean that might not be huge).
This understanding was informed by recently reading "Not the end of the world" by Hannah Ritchie, but I'm open to contradicting evidence.
Yes I know about the argument of reducing farmland, supposedly up to 40% of it is used for animal feed.
I'm already very skeptical of such high figures, to me it sounds like the absurdist argument of cow water use, that include all the water used for the lifecycle of the animal and more while conveniently forgetting that we use a ton of water to process plant both at the exploitation stage and consumption stage (good luck consuming cereals and legumes without water). To me it sounds like a reductionist argument that tries to compare things that really aren't comparable.
But even if we accept the premise, it seems based on a fantasy world where every land is equal to another and you can produce any kind of crop anywhere. Animal feeds are generally low-quality crops that farmers put on soil where the good stuff doesn't grow that well. It's also used for cycling, when soil is too depleted to support another growth with decent yields.
Even if we would stop all animal farming overnight, the amount of land you could reclaim for plant agriculture would never match what was freed. If you look where they do intensive animal farming, it tends to be in places where plant farming would be very difficult at the very least. One easy example is the milk cows in the mountains.
I go running in semi-wild space near a river and I often find grazing cows in the swamps. A lot of the land near the river is actually used to graze cows or to grow hay that will feed the cows in the winter. You can't really use the land for anything else really, one year the farmer tried putting wheat and he never did it again. I suspect the yield and quality wasn't good enough, so it was mostly a waste of ressources. However, another nearby farmer has been farming corn, which is definitely used as animal feed. The crop is probably more tolerant to the environmental pressure of this land and has good yields (has been that way for over 20 years).
What's more, the thing is that animal farming fundamentally doesn't get you the same stuff as plant farming. The macro nutrient ratio is not the same and even the micro nutrient profile is very different. You can't say I'll substitute beef with some random cereals; they are definitely not the same.
So, you need different crops, like legume, that are much more demanding on the soil and don't grow as well everywhere as the common cereals.
Potentially you could re-arrange things around and optimise to get the best yield for every type of plant depending on what's needed at the level of a country for example. But this is an optimisation problem and those are already very hard but with many independent actors who will try to maximise their profits it's almost impossible. I guess we could go full on communism but I think you can understand why that would be even more undesirable than animal farming.
On top of that, animals are very often eating waste byproducts of plant farming. One example is cattle cake, derived from soybean oil production. This is just a single example and from what I coud find, it's not an insignificant amount of waste that gets "recycled" like that. It's not clear that we could do anything else with it, so in a way animal farming has a virtuous component where it "upcycles" waste.
I'll ask a question: if plant farming is so much more efficient, how come things like soya steaks are still expensive? They are not cheaper than most animal proteins, at every comparison point (weight, protein ratio, caloric density). It doesn't make any sense. Logically if getting proteins from plants is really more efficient than animals, they should be much cheaper, that's basic economics.
I believe this is because most arguments around plant farming do not factor everything in the equation and that makes plants look better. You can eat a steak with minimum preparation and energy use (technically you can eat it raw if you are sure there is no contamination risk), throw it in a pan for a few minutes and it's done. But plants like soya need a lot of processing, using both a lot of water and energy. Even if you get the raw stuff, it needs to be washed and cooked, and those things take a very long time. I make hummus from raw chickpeas very often and cooking time is at least 45 min. Suddenly you need to add energy cost, water cost at the very least (and time cost, but we can try to ignore it and pretend we are all money poor but time rich). In similar fashion, tofu needs a shit ton of water and lots of energy for cooking (they actually have problem with waste management from tofu production in Asia because the water used is often released directly in rivers and it kills the fishes).
I think that if plant food was that much more efficient, it would already be reflected in the consumer price. You could attribute high prices to greed and low volume but that doesn't make a lot of sense. If it was possible, producers would undercut to get larger parts of the market even if it is small compared to animal products (it is still a big market at the country level so there are definitely profits to be made if that was possible).
For commodities like food, I think the price reflects the efficiency of ressource use to get the product to market, it's the concept of embodied energy. If plant-based products don't do better it's probably because they aren't actually any less wasteful than animal farming.
After you have considered those things, it is necessary to consider the impact on health/feeling. There are plenty of people who have tried veganism and couldn't stick to it for health-related reasons. Even if you are not technically sick, feeling good is not something most people would like to give up. Surviving is one thing but you'll have a hard time convincing people to give up animal foods if they end up feeling like an inmate in a concentration camp.
This is my experience. I have a very chaotic diet for many reasons and sometimes I "forget" to eat animal food. I become an unplanned vegetarian/vegan from time to time for a few days. When I start to feel weird and think about eating some meat, the feeling on the subsequent day is really incomparable. I suspect this is the same experience that many ex-vegans tell, and it is hard to handwave away. What's the point of protecting the environment, if your life ends up being miserable when doing so? It's a very hard sell and I think that short of a complete ban on animal products, people will never give them up.
Even if all of this was completely wrong, it seems preposterous to focus on diet as a way to reduce environnemental damage. Eating is a fundamental experience of life and it isn't just the thing that gets you to survive, it's a pleasurable thing and a social activity as well. Coming up with moral arguments to justify how people should eat is akin to religious behavior and unsurprisingly all religions have all kinds of diet requirements. It doesn't seem reasonable to completely rework diet solely to reduce environmental damage (it's not even clear how much reduction would be really possible in the first place).
This is particularly true when there are many other things that we could do to reduce environnemental damage and wouldn't touch one of the fundamentals of human life. For example, the overuse of the single ownership car design lifestyle, abuse of international travel purely for leisure and in fact plenty of things we do solely for leisure or in social status. Or things like purchasing all kinds of crap we don't need and really most of the stuff of the capitalist economy where people buy stuff to use it a few times at best when it could be shared by many.
Before requiring people to change their diet we could heavily tax consumer products coming from countries with very lax environmental laws and from companies that make things that don't last, fast fashion, etc...
All the arguments around vegetarianism/veganism always seems like virtue signaling and a cheap attempt at getting moral high ground. If one has to make an effort to reduce environmental harm, there are plenty of low hanging fruits of lifestyle change before having to touch at the diet.
For this reason, even if all the propaganda around the plant-based farming/diet turns out to be completely true, I don't think it matters all that much.
That being said, I am still very much interested in the correct answer, I just want to know for sure, so I'll keep reading on the subject.
Personally I already have a low impact lifestyle so I don't feel like giving up animal food on top of that, especially since it makes me feel like shit...
Sorry it took a while to reply, I wanted to respond properly. I also appreciate you taking the time to engage in this.
Unfortunately, the 40 percent figure for cropland used for animal feed isn’t propaganda; it’s consistent across FAO and peer-reviewed studies. The consumption argument you make (that we can’t consume those crops without water) also applies to animals. As you move further up the food chain you’re just multiplying inefficiencies.
When it comes to land use, you are right that some pasture land can’t grow crops. If the demand for meat limited us solely to grazing on those lands, things would be better. The issue is the demand is far greater: we are literally cutting down the Amazon rainforest to make room for grazing. Prime arable land, especially maize and soy, is also being used to feed animals rather than humans.
The price doesn’t reflect efficiency either. Meat is very heavily subsidised in the West and in Asia. It also benefits from massive economies of scale and externalises huge costs like emissions, water pollution, and antibiotic resistance. In any case, here in Europe tofu is already far cheaper than meat. See this: one euro (or dollar?) for 24 g of protein, hard to get more efficient than that: https://www.dm.de/dmbio-tofu-natur-p4067796251999.html
It’s also misleading to compare beef to cereals in isolation. Balanced plant-based diets with legumes, nuts, and vegetables are nutritionally adequate and offer a much richer microbiome diversity when supported with B12. There are reports of people feeling worse on vegan diets but normally that is because they aren’t eating a variety of fruits and vegetables but relying on processed stuff. Meat makes it easy to ‘cheat’ your way to getting all the nutrients you need in the short term. That works great until you have a coronary heart attack in your 50s.
Food systems are responsible for roughly a quarter of global greenhouse gas emissions, with livestock accounting for about half of that. Many independent studies support this. Peer reviewed studies such as this one by Oxford, taking into account all the factors you mention above, also suggest vegan food emissions are 30% of those of meat eaters: https://pubmed.ncbi.nlm.nih.gov/37474804
'All the arguments around vegetarianism/veganism always seems like virtue signaling and a cheap attempt at getting moral high ground' - To this I would just say, I was a meat eater for more than 30 years. I also found vegans 'annoying'. But if you really look into it the evidence is overwhelming. That's even before you take into account the fact that most vegans (such as myself) are doing it because of the massive suffering we are inflicting on living, sentient beings.
I find it ironic that we pet animals, or discuss here about AI/superintelligence ethics, when we literally torture animals in slaughterhouses. I doubt anyone here could tour a slaughterhouse and come out a meat eater.
So where is the real propaganda? From vegans? Or from the meat industry who force policymakers to rename 'Oat Milk' to 'Oat Drink'? Who subsidise the meat industry with vast sums? Who do everything in their power to hide the horrors of slaughterhouses from the public (even prosecuting people filming inside these facilities for 'defamation' here in Germany, and winning - even though they are literally just showing what happens inside). The so-called “free-range” or “happy cows” marketing is also propaganda: animals spend a few days outside a year, have their young taken from them, and are then transported long distances to be killed, often in extreme distress.
Anyway, I got a bit carried away. It's good to discuss this stuff. I can recommend the book 'How to Love Animals' by the FT journalist Henry Mance, it was quite an important read for me. On general dietary topics (vegan or not), and eating well on a plant based diet, I can suggest anything by Tim Spector and his recommendation to get 30 types of plants a week in one's diet.
>I didn't realize it for 30 years, but it's hard to feel connected to nature, animals, and the environment when you are eating something you didn't kill yourself.
Really? I'd think that living surrounded by a modern society whose benefits you fully enjoy does a lot more to really disconnect you from nature than some notion of not killing the meat you ate.
You're still fully participating in the daily destruction of nature, animals and living things just about as much as anyone who eats meat, you've simply removed yourself symbolically a bit more from one specific expression of it, so you can (entirely subjectively) feel as if it somehow makes much of a difference for any real connection to the planet.
"You're still fully participating in the daily destruction of nature, animals and living things just about as much as anyone who eats meat" - I mean, that's categorically not true. All else equal the meat eater has a far more destructive daily impact on nature.
I can also pet a dog with a good conscience, because I don’t turn around and eat an animal just like it. I don’t see one as a friend and the other as food just because society dictates that
Well, the animal died in an intensely stressful situation that you would probably struggle to even watch, let alone do yourself. I guess all that cortisol and stress is now also part of you.
Vegans are just as disconnected, especially when they eat things grown in vats. What brings connection is to have a garden and kill and eat the animals and vegetables they grow there.
The point is that you're not relegating another sentient being's life to lower than that of your own. You refuse to accept the torture and murder of another species.
In doing so, you dramatically expand your empathy and understanding of what it is to be a living thing, and hence you gain an inner connection to nature and animals that is hard to describe. At least that's my personal experience.
You are still relegating other life you eat as lower than yours, just not the sentient ones. Also there are the animals that die horrible deaths during machine harvesting. The modern monoculture cropping taking away habitat from nature, and destroying land in the process. Theres also the unethical food transport to get it to you as well.
You can be empathetic about life and still recognise your biology and your place in the ecosystem. Usually cultures around the world celebrated this fact by being thankful and mindful of the food in front of them, especially if you yourself have slaughtered the animal.
One should be free to avoid eating the meat of sentient life; but one should also be free to accept that their body functions best by eating like the apex predator they are. Seeing the world in black-and-white is what causes all the ills in the world: being human is accepting that we are a walking bag of paradoxes, and perfection is unattainable.
The lack of judgement is why I personally admire self-described vegetarians much more than militant vegans, always trying to convert the world to their righteous image. Live and let live.
If we are all apex predators, then why do we have laws? Why can't the strongest person attack someone and take their money? Why can't someone kill a dog and eat it?
Why don't we just poop wherever we want, for that matter?
Because we are more than animals. We can be better. We must be better.
Ironic to use 'live and let live' in that context.
The same week Oracle is forecasting huge data center demand and the stock is rallying. If these 10x gains in efficiency hold true then this could lead to a lot less demand for Nvidia, Oracle, Coreweave etc
Sure but where is the demand going to come from? LLMs are already in every google search, in Whatsapp/Messenger, throughout Google workspace, Notion, Slack, etc. ChatGPT already has a billion users.
Plus penetration is already very high in the areas where they are objectively useful: programming, customer care etc. I just don't see where the 100-1000x demand comes from to offset this. Would be happy to hear other views.
We are nearly infinitely far away from saturating compute demand for inference.
Case in point; I'd like something that realtime assesses all the sensors and API endpoints of stuff in my home and as needed bubbles up summaries, diaries, and emergency alerts. Right now that's probably a single H200, and well out of my "value range". The number of people in the world that do this now at scale is almost certainly less than 50k.
If that inference cost went to 1%, then a) I'd be willing to pay it, and b) there'd be enough of a market that a company could make money integrating a bunch of tech into a simple deployable stack, and therefore c) a lot more people would want it, likely enough to drive more than 50k H200s worth of inference demand.
Do you really need a H200 for this? Seems like something a consumer GPU could do. Smaller models might be ideal [0] as they don't require extensive world knowledge and are much more cost efficient/faster.
absolutely nobody wants or needs a fucking thermostat diary lmao, and the few ppl that do will have zero noticeable impact on world's compute demands, i'm begging ppl in on hn to touch grass or speak to an average person every now and then lol
You wouldn't even know that it existed, or how it worked. It would just work. Everybody wants hands off control that they don't have to think or learn about.
edit: this reminds me of a state agency I once worked for who fired their only IT guy after they moved offices, because the servers were running just fine without him. It was a Kafkaesque trauma for him for a moment, but a massive raise a week later when they were renegotiating for him to come back.
its pretty easy to dispute and dismiss a single use case for indiscriminate/excessive use of inference to achieve some goal, as you have done here, but its hard to dispute every possible use case
You will ALWAYS want to use the absolute best model, because your time is more valuable than the machine's. If the machine gets faster or more capable, your value has jumped proportionally.
> Plus penetration is already very high in the areas where they are objectively useful: programming, customer care etc.
Is that true? BLS estimates of customer service reps in the US is 2.8M (https://www.bls.gov/oes/2023/may/oes434051.htm), and while I'll grant that's from 2023, I would wager a lot that the number is still above 2M. Similarly, the overwhelming majority of software developers haven't lost their jobs to AI.
A sufficiently advanced LLM will be able to replace most, if not all of those people. Penetration into those areas is very low right now relative to where it could be.
Fair point - although there are already so many customer facing chatbots using LLMs rolled out already. Zendesk, Intercom, Hubspot, Salesforce service cloud all have AI features built into their workflows. I wouldn't say penetration is near the peak but it's also not early stage at this point.
In any case, AI is not capable of fully replacing customer care. It will make it more efficient but the non-deterministic nature of LLMs mean that they need to be supervised for complex cases.
Besides, I still think even the inference demand for customer care or programming will be small in the grand scheme of things. EVERY Google search (and probably every gmail email) is already passed through an LLM - the demand for that alone is immense.
I'm not saying demand won't increase, I just don't see how demand increases so much that it offsets the efficiency gains to such an extent that Oracle etc are planning tens or hundreds of times the need for compute in the next couple of years. Or at least I am skeptical of it to say the least.
We've seen several orders of magnitude improvements in cpus over the years, yet you try to do anything now and interaction is often slower than that on zx spectrum. We can easily fill in order of magnitude improvement and that's only going to create more demand. We can/will have models thinking for us all the time, in parallel and bother us with findings/final solutions only. There is no limit here really.
I’m already throughput-capped on my output via Claude. If you gave me 10x the token/s I’d ship at least twice as much value (at good-enough for the business quality, to be clear).
There are plenty of usecases where the models are not smart enough to solve the problem yet, but there is very obviously a lot of value available to be harvested from maturing and scaling out just the models we already have.
Concretely, the $200/mo and $2k/ mo offerings will be adopted by more prosumer and professional users as the product experience becomes more mature.
The difference in usefulness between ChatGPT free and ChatGPT Pro is significant. Turning up compute for each embedded usage of LLM inference will be a valid path forward for years.
The problem is that unless you have efficiency improvements that radically alter the shape of the compute vs smartness curve, more efficient compute translates to much smarter compute at worse efficiency.
Isn't that essentially how the MoE models already work? Besides, if that were infinitely scalable, wouldn't we have a subset of super-smart models already at very high cost?
Besides, this would only apply for very few use cases. For a lot of basic customer care work, programming, quick research, I would say LLMs are already quite good without running it 100X.
MoE models are pretty poorly named since all the "experts" are "the same". They're probably better described as "sparse activation" models. MoE implies some sort of "heterogenous experts" that a "thalamus router" is trained to use, but that's not how they work.
> if that were infinitely scalable, wouldn't we have a subset of super-smart models already at very high cost
The compute/intelligence curve is not a straight line. It's probably more a curve that saturates, at like 70% of human intelligence. More compute still means more intelligence. But you'll never reach 100% human intelligence. It saturates way below that.
Thanks, I wasn't aware of that. Still - why isn't there a super expensive OpenAI model that uses 1,000 experts and comes up with way better answers? Technically that would be possible to build today. I imagine it just doesn't deliver dramatically better results.
I mean 640KB should be enough for anyone too but here we are. Assuming LLMs fulfill the expected vision, they will be in everything and everywhere. Think about how much the internet has permeated everyday life. Even my freaking toothbrush has WiFi now! 1000x demand is likely several orders of magnitude too low in terms of the potential demand (again, assuming LLMs deliver on the promise).
I'm not going to speculate about what might be ahead in regards to Oracle's forecasting of data center demand, but regarding the idea of efficiency gains leading to lower demand, don't you think something like Jevons paradox might apply here?
People said the same thing for deepseek-r1, and nothing changed.
If you come up with a way to make the current generation of models 10x more efficient, then everyone just moves to train a 10x bigger model. There isn’t a size of model where the players are going to be satisfied at and not go 10x bigger. Not as long as scaling still pays off (and it does today).
Absolutely not; the trends have proven that people will just pay for the best quality they can get, and keep paying roughly the same money.
Every time a new model is released, people abandon the old, lower quality model (even when it’s priced less), and instead prefer to pay the same for a better model.
Sure but the money people are paying right now isn't that much in the grand scheme of things. OpenAI is expecting 13bn in revenue this year. AWS made over 100bn last year. So unless they pay a lot more, or they find customers outside of programmers, designers, etc who are willing to pay for the best quality, I don't see how it grows as fast as it needs to (I'm not saying it won't increase, just not at the rate expected by the data center providers)
For early adopters yes but many systems have been running as good enough without any kind of updates for a long time.
For many use cases it needs to get to a point where accuracy is good enough and then it will be set and forget. I disagree with the approach but that's what you find in the wild.
The best quality you can get is at odds with the best speed you can get. There are lots of people (especially with specific use cases) who will pay for the best speed they can get that is high enough quality.
If someone had to bet on an AI crash which I imagine would led to unused datacentres and cheap GPUs how would they invest their winnings to exploit these resources?
If the price of inference drops through the floor all the AI wrapper companies become instantly more valuable. Cursor is living on borrowed time because their agents suck and they're coasting on first mover advantage with weak products in general, but their position would get much better with cheap inference.
No. The gains in inference and training efficiency are going to be absorbed by frontier LLM labs being more willing to push more demanding and capable models to the end users, increase reasoning token budgets, etc.
For the last 2 years, despite all efficiency gains, I am literally watching characters appear on my screen, as if this was a hacker movie. Lately, I am also waiting for at least 60s for anything to appear at all.
If that happened at 10x the speed, I would still be slow in computer terms, and that increasingly matter, because I will not be the one reading the stuff – it will be other computers. I think looking back a few years from now, every single piece of silicon that is planned right will look like a laudable but laughable drop in the ocean.
The real quality demand needs is not there, so more processing is very probably needed, so efficiency gains may allow the extra processing.
(A string example read today of Real quality demand needs: the administration of Albania wants some sort of automated Cabinet Minister. Not just an impartial and incorruptible algorithm (what we normally try to do with deterministic computation): a "minister". Good luck with that.)
Even that is false. There are many more countries with fewer scams per capita, and with fewer amount lost at scams per capita. Meaning, Singapore falls behind other countries in elimination of scams.
