For the sake of argument, what if Amazon decided tomorrow that they would secure exclusive contracts with all food suppliers and then hoard all the food to starve out the people they don't want to have it? Or at least, drive up the price of food so it becomes completely unaffordable? I know people can simply grow their own food so it's a bit different, but hopefully it gets the point across. It's anti-trust on an unprecedented level.
But OpenAI legitimately needs HBM. Amazon in this instance doesn't need food and is doing purely to create artificial scarcity. If OpenAI were to actually not use the HBM then it could mean something.
That's the whole problem: it's unlikely that OpenAI will actually use all of that HBM. It seems probable that they are using it to create artificial scarcity for their competitors.
IANAL, but yes, I believe it can nullify the claim. Bumping into someone on the sidewalk is only battery if the prosecution demonstrates intent to harm.
Yeah it would definitely still be a problem, but history shows that life finds a way. Even if everyone has to eat nothing but planted potatoes from any patch of grass that one can lay eyes on.
What history has taught us is that life finds a way by staying together and each person having their function within society, only some of which is growing or producing food.
Surprisingly, many do. When I mention to people (family, friends, etc.) that they should open new chat windows for new topics due to memory corruption, it's pretty clear they never even considered the possibility that the model can go off the rails if the chat is too long. Later I often get a comment like "wow, I kept thinking this AI stuff was rubbish, but it's really good now!".
I suspect that the "image-based" strategy taken here is unlikely to be appealing to many members of this community.
It could be very effective for bringing in those who are not particularly computer literate under the claimed guarantee that a random update is unlikely to break the machine. But you would also need significant financial backing and marketing with strong brand recognition to inspire that kind of confidence.
I think the right way to do this is with snapshots, the way opensuse microos is doing it, for example. You get the best of both worlds that way - you still can easily install packages into the OS to customise it, and you do get painless updates and rollbacks. There's a very narrow use case where you _do_ want images, but for that you'll want to control the complete secure boot chain for attestation, so I'd dismiss it here.
Fun fact, a bit over a decade ago we were probably the first one ever to publish a distribution to rely on btrfs snapshots per default with the Jolla phone. Sadly that did bite us due to reliability of btrfs at the time, and later phones switched to ext4, but with a stable filesystem it's a nice mechanism for handling updates and factory reset.
I would think of my self as atleast computer literate and i very much prefer atomic linux to traditional distros, arch or nixos. If you are in luck with hardware - you get system that is hard to polute with my actions and everything developer i do in separate distrobox. Rolling back versions or even hoping to completely different immutable distro is just restart away. I've never been so peaceful with linux.
I don't buy it. It's easier to make this argument for companies that are building their own hardware, since they know it can be immediately used. OpenAI's move is tantamount to hoarding for the sake of strangling competition. There was plenty of supply to allow for their plans without this move (especially since they will probably go bankrupt at this rate).
If stored in proper conditions, the shelf life of undiced wafers is pretty much limitless. But if moisture or dust get in, it’ll start to have corrosion and other damage. Diced wafers on the other hand are placed on UV tape and I recently learnt that the shelf life of that is like a few months at best.
Nevertheless, unless you’re not storing them in a clean room of appropriate class (or in a vacuum pack), assume 18 months for undiced wafers.
In terms of size, I would think that stored in cassettes or carriers a vacuum sealed, it shouldn’t be too much. A 25 wafer pack is about half meter cube? A 40ft shipping container is about 68m^3. So one container can store about 140 packs of 25 wafers each. That’s 3500 wafers. They’re talking about 900k wafers. That’s about 260 containers. Call it 300 with extra space and stuff. Not exactly hard to provision.
Granting this premise is true (I have no idea), that makes it even worse. They would deliberately be hoarding 40% of the global supply not to lock in for future growth but simply to make sure no one else gets to have it. It’s figuratively setting chips on fire.
As an AI researcher, I thought it was relatively well established (at least among my colleagues) that being pro-AI actually meant you were anti-Sam as well. He's the worst actor in the industry and has done an incredible amount of damage to its brand.
We left behind any pretense of a free global market once we entered a post-tariff world. You can't have large universal tariffs or even the threat of them and expect the market to act freely, the two are fundamentally incompatible.
There has always been pressure to do so, but there are fundamental bottlenecks in performance when it comes to model size.
What I can think of is that there may be a push toward training for exclusively search-based rewards so that the model isn't required to compress a large proportion of the internet into their weights. But this is likely to be much slower and come with initial performance costs that frontier model developers will not want to incur.
> so that the model isn't required to compress a large proportion of the internet into their weights.
The knowledge compressed into an LLM is a byproduct of training, not a goal. Training on internet data teaches the model to talk at all. The knowledge and ability to speak are intertwined.
I wonder if this maintains the natural language capabilities which are what LLM's magic to me. There is a probably some middle ground, but not having to know what expressions, or idiomatic speech an LLM will understand is really powerful from a user experience point of view.
I don't see this working for Google though, since they make their own custom hardware in the form of the TPUs. Unless those designs include components that are also susceptible?
That was why OpenAI went after the wafers, not the finished products. By buying up the supply of the raw materials they bottleneck everybody, even unrelated fields. It's the kind of move that requires a true asshole to pull off, knowing it will give your company an advantage but screw up life for literally billions of people at the same time.
Still susceptible, TPUs need DRAM dies just as much as anything else that needs to process data. I think they use some form of HBM, so they basically have to compete alongside the DDR supply chain.
(1) I always tell my students that if they don't understand why things are done a certain way, that they should try to do it in the way most natural to them and then iterate to improve it. Eventually they will settle on something very similar to most common practice.
(2) Higher-level proofs are using so many ideas simultaneously that doing this would be tantamount to writing Lean code from scratch: painful.
For the sake of argument, what if Amazon decided tomorrow that they would secure exclusive contracts with all food suppliers and then hoard all the food to starve out the people they don't want to have it? Or at least, drive up the price of food so it becomes completely unaffordable? I know people can simply grow their own food so it's a bit different, but hopefully it gets the point across. It's anti-trust on an unprecedented level.
reply