Thought I could wait to build a new PC, but at this point, it’ll be another 5-10 years before things start getting better. After the price of the steam machine, I ordered parts to build my own. Such a bad time to build, but it’s just gonna get worse the longer I wait.
How long does it take a company already established in the sector to build a RAM factory?
Around 4 years. The creation of any microchip requires extremely stable geology, pristine rooms, high levels of training, Scrooge McDuck moneybins, and so on. It isn’t an industry that can turn on a dime.
Things are probably going to get worse in the short term, but the AI bubble is going to burst. Magnitude? We’ll see, but when investors realize that these companies cannot make a profit, and open source frontier models that allow you to run AI in house are removing vendor lock in, things are going to change. Also, LLMs are a dead end, and have little room to improve.
Newer paradigms are appearing, such as Yann LeCun’s JASP, which actually learns, and other approaches, which will make LLMs obsolete, and are way less hardware intensive.
Another factor is the Chinese closing in in consumer grade RAM. If it can be proven that no backdoor or other shenanigans are there, they will balance things somewhat.
While current reality is what it is, there may be a massive social and traditional media manipulation by the big three and other interested parties to fuel fear of rising prices forever, to push people to buy as much as they can at these prices. I have no proof of this, but I don’t think it’s far fetched.
And let’s not forget that for media outlets, fear and tragedy sells. (I think Hearst or some other news mogul said that last century.)
Yann LeCun’s JASP
Wait, this is my first time reading about this. Got an ELI5 or TL;DR?
Courtesy of Kagi’s search AI:
000000000
It appears there is a slight misunderstanding of the acronym: Yann LeCun’s architecture is called JEPA (Joint Embedding Predictive Architecture), not JASP.
JEPA differs from Large Language Models (LLMs) primarily in how they learn, what they predict, and how they represent the world. While LLMs are “word models” designed to generate sequences of tokens, JEPA is intended to create “world models” that understand the underlying physics and logic of reality 3 4 .
The key differences are summarized below:
-
Generative vs. Predictive (Non-Generative)
LLMs are Generative: They operate by predicting the next token in a sequence (generative AI) 2 . This approach often leads to hallucinations because the model focuses on statistical probability rather than factual ground truth 6 . JEPA is Predictive: Instead of generating every single pixel or word, JEPA predicts latent representations (embeddings) in a hidden space 5 . It tries to learn what is “plausible” rather than attempting to reconstruct every single detail of the input.
-
Word Models vs. World Models
LLMs are “Word Models”: They learn from text and treat intelligence as a language manipulation task 4 . LeCun argues that language captures only a small subset of human thinking and cannot represent high-dimensional physical spaces 2 7 . JEPA aims for “World Models”: It is designed to understand cause and effect, physics, and the physical environment 1 . This allows the system to reason from first principles and plan sequences of actions, which is a prerequisite for autonomous AI 1 .
-
System 1 vs. System 2 Thinking
LLMs (System 1): LeCun describes LLMs as “System 1” processes—they are reactive and perform a fixed amount of computation to produce each token 2 . JEPA (Path to System 2): By incorporating world models, JEPA is intended to enable “System 2” thinking—the ability to plan, reason, and deliberate before acting 1 .
Summary Comparison Table Feature Large Language Models (LLMs) JEPA / World Models Core Goal Predict next token (Text/Code) Predict latent state (Reality/Physics) Method Generative (Pixel/Token by pixel/token) Joint Embedding (Non-generative) Domain Linguistic/Statistical patterns Physical/Causal understanding Weakness Hallucinations, lacks physical grounding Limited fluency in natural language Cognition Reactive (System 1) Planning/Reasoning (System 2)
-
I’ve come to terms with the fact I’m not going to be buying any new computer-type devices until the bubble pops.
I’m just terrified what happens if one of my existing devices breaks. If a RAM stick goes bad, I might have to mortgage my non-existent house.
In case you are being serious, there are ways around faulty RAM sticks, usually just a few cells/rows are affected. In case anyone needs to know this, here is a pretty good summary on stackoverflow on how to deal with this on linux. In general, look for “memmap”.
Keep your hardware running as long as possible! Iirc newer RAM is unfortunately somewhat more susceptible to failing. My DDR3 is still working fine.
Time to hit the books, explore the mountains, build a chicken shed, grow something in the garden, buy a plot of land and grow something… Build a house myself idk things like that.
Yep, definitely not buying any new computers any time soon. If anything breaks I’ll try to live with it as much as I can.
Fuck this pricing.
I got a framework some time ago, best decision ever. I can upgrade anything and everything on my laptop when I want, how I want, where I want.
That doesn’t make those components magically cheaper…
I can upgrade anything in my desktop too, but the 4x32gb ddr5 sticks I paid $200 for in 2024 now cost around $1,600.
Ow yeah 100%, just A case for reusability, repairability, and longevity. This might help users to extend the life of their computers, thus long term lower costs for the user itself when extending the life of the machine.
You may hate AI, but it’s not the reason we are seeing RAM costs skyrocket.
Looking at the manufacturing data and the historical strong-arm tactics used by Samsung, Micron, and SK hynix, who collectively control about 96% of the global DRAM market, AI just gives manufacturers the perfect public justification to stop chasing cheap bit growth, starve low-margin consumer channels (our RAM products), reprioritize wafers toward premium products (data center RAM), and force customers into multi-year contracts at shortage-era prices.
Sure it lets them make more money off us, but they really love locking in these rates with their data center-based customers.
And all this is just because … AI demand.
Rather they grabbed a fancy reason to increase prices because of greed.
The few big ram production companies likely had a chat and decided: “Hey, more money? Yea! More money! We like more money!”
But of course they do, their shareholders wouldn’t like it otherwise. Capitalism 101. 🤷
it will be ironic when consumers cant afford the end user devices needed to interact with the AI servers. House of cards??
Let me put on my tinfoil hat really quick.
They want to kill personal computing. You don’t need a full blown computer, you need a fire stick style device that plugs into your monitor and allows you to remotely access the virtual machine you rent from Microsoft on a monthly basis.
There is no tinfoil about it. Jensen Huang and the other tech oligarchs have openly stated that is their goal.
this has been “the plan” since 2005. when i was in high school trying to figure out what the heck “cloud computing” was, this is what they were talking about: anything requiring more compute than secure authentication and pixel drawing would be rendered in the cloud and delivered to dumb terminals. this is what netbooks, Chromebooks, and smartphones have been a step towards if not an implementation of.
Looking back at computing history, cloud computing is basically reverting back to the original mainframes and dedicated terminals.
There was a hype of using thin clients, the concept is that you get just enough hardware and software to be able to connect to a session running on a shared server, the admin can allocate more resources like CPU cores, RAM and storage as individual needs change over time.
As an IT guy, I do like the concept in a corporate environment, especially when looking at the SunRay system from Sun, which used smartcards for easier access, you put your card into the client and if configured properly, you got your old session loaded and ready in a few sec, regardless of which client you put your card into.
The YT channel Clabretro has several interesting SunRay videos.
Sun ray was cool.
We had them at university with smart cards.
Yup, there’s a cycle between centralization and edge. Started with centralization, mainframes, went edge with the first PCs (and game consoles) and ever since corpos have been trying to pull it back to the center in waves. Thin client, cloud compute, arguably phones (as apps processing in the cloud), Geforce Now, AI. So far it’s always gone back to the edge for most of the population, except for niche cases (or not in the case of phones). As internet gets faster and more reliable the chance of it sticking longer in the central zone increases (IMO).
As internet gets faster and more reliable the chance of it sticking longer in the central zone increases
It also allow the opposite, the fediverse is a living example of it, the more internet is faster the more we can organize better to decentralize, it’s harder that someone get in contact with the decentralized internet but it’s more probable that it happen now than 7 years ago
Some companies also would still continue to sell consumer hardware (like Framework, probably)
I see it as a fight, companies are currently winning sadly but we have the power to win too
Heartily agree. I was trying for objectivity, personally you’ll take my general purpose computers from my cold dead hands, and I will encourage everyone I can to support it.
I take some solace from the general pendulum nature of the process (and societal evolution in general), this too shall pass. Doesn’t mean we don’t need to fight. Evil triumphs when good people do nothing.
As an end user with this model at work, I hate this slow-ass shit that keeps freezing on me and randomly disconnects. Not fun or productive when it starts talking seconds to register a click because the connection becomes unstable for no apparent reason.
Wait, are you still running a SunRay at work?
Wow, I thought they were all scrapped
Man, this makes me want to take stock of what I have in the attic. Those old computers dont sound so bad any more.
I also wonder if I should be hoarding raspberry pi and building some kind of cluster
Even those contain RAM.
Yes, but they can get away with a tiny amount of it. Perhaps we are back at the start of personal computing and arcade machines era, where every byte counted.
Dumb terminals don’t need much RAM. Unfortunately, the minimal RAM would come with maximum rentiership and exploitation.
i am having visions of 3270 screens😉
I see this as an opportunity to get invested in non-tech hobbies because there’s nothing else I can do
Yup, was going DDR5 this year, got a new bike (Marin Larkspur) instead, such a good decision, renewed my love of cycling, having a ball.
Books my man, books.
Mini painting, get addicted to plastic crack. Seems to be cheaper than tech hobbies these days.
Get a resin printer and you can print infinite minis.
That would probably be too close to a tech hobby and considering what some of the tech giants are doing. 3D printing will be labeled under criminal activity
Already happening. Those printed gun laws that are being pushed in multiple states are not about guns, they’re an assault on small scale manufacturing and the right to self repair.
The RAM I bought in 2019 for $100 is now over $500. at this rate I’ll sell it in a few years to put my daughter through college.
These price gouging assholes should probably be prosecuted along with the “AI” fraudsters.
At this rate I’ll be able to sell my ddr3 from over a decade ago and make a profit compared to what I paid originally
It’s a good thing I saved the 24GB of DDR3-1600 from my old laptop. I might actually be able to get some money for it.
I have 256GB DDR3 ECC, I’m ready for RAM to turn into 2017 bitcoin at this point!
No reference to CXMT’s memory that’s just starting to be sold in western markets now, I’m very curious if/how that’ll affect things. If corsair’s cxmt kit can hold pricing steady, let alone drop it, for a 2x16 6000MHz CL36 kit then it could be huge
Yeah there’s a lot of doom here, but I keep saying that there are reasons that they are locking in 5 year price guarantees, it’s because they know it’s going to crash back down, whether it be bubble or normalization or Chinese. It’ll take some time, but it’s not forever.
I’m really glad I got my steam deck last year when I had the chance.
8Gb of ddr4 is $25 usd and 16Gb ddr4 sticks are going for about $75 usd goes up from there depending on speed.
8Gb of ddr4 is $25
Ha! My 32g was $600, so 6x that.
I just bought a 32GB SODIMM DDR5 5600MHz for $430 😅 at least it was for a work PC
