It is of course true, as the saying goes, that the market can stay irrational longer than you can stay solvent.
They’re giving OpenAI a subsidy for chips that OpenAI would have to buy anyway. Nvidia has no real competitors in this segment. If the “AI economy” is as big as companies like OpenAI claim it is, it doesn’t need Nvidia’s help.
OpenAI doesn’t have $100 billion to invest. That’s why they sold equity to Nvidia. (It also jacked their valuation.)
No, that’s the whole point - Nvidia is ensuring there will be a competitive AI economy, by investing in OpenAI’s growth. It’s stoking the fires of competition by making it possible for OpenAI to dominate. Microsoft, Google, X and several Chinese companies will need to escalate to stay relevant, and the escalation requires buying more processors from Nvidia.
It might be helpful to take a step back and look at the bigger picture. Right now, the AI hardware race is all about GPUs—and NVIDIA owns that space. But the emerging frontier is quantum computing, especially for AI workloads that push past classical limits (as I suggested in an earlier post).
NVIDIA sees that shift coming and isn’t sitting still. It’s not building quantum chips but is investing heavily in the ecosystem by partnering with quantum hardware startups, building hybrid platforms like CUDA-Q to link quantum processors with its GPU supercomputers, and even launching a dedicated quantum research center. Basically, it’s positioning itself as the bridge between today’s AI acceleration and rapidly emerging quantum breakthroughs. When quantum hits prime time, NVIDIA wants to be the one holding the keys to integration.
Given that, I do not think NVIDIA’s strategic relationship with OpenAI is about selling more GPUs but rather a play to ensure it effectively controls the AI component of the emerging quantum computing/AI market it’s working to build and control.
I don’t think it’s that at all. Neither OpenAI or Nvidia have any interest at all in “stoking competition”. That’s just now how it works now in big tech. They want to own the market, and bury everyone else.
I find it fascinating to watch for-profit titans jockey to control and benefit from OpenAI’s work in AI, given that OpenAI is a 501(c)(3) nonprofit and must operate in the public benefit rather than for private profit.
To accomplish this matter/antimatter sleight of hand, in 2019 OpenAI spun up a for‑profit arm called OpenAI LP. The nonprofit sits at the top as the governing body, while the LP is where outside investment occurs. That’s how companies like Microsoft—and now NVIDIA—can take equity stakes. The twist is that it’s a ‘capped‑profit’ model: investors can make returns, but only up to a set multiple, after which any upside flows back toward the nonprofit mission. So the nonprofit keeps its tax‑exempt status, the for‑profit raises the billions needed for compute and talent, and investors get limited but still potentially huge returns. In other words, the nonprofit holds the steering wheel, while the LP is the engine that takes in fuel from private capital.
At first, it occurred to me that, if it were incorporated in the state where I practiced for decades, California, there would be a serious risk that the California Attorney General would collapse all of this down and see it as a sham (which, of course, it is). However, it’s incorporated in Delaware, where the state’s Attorney General has no oversight, so state-imposed revocation of tax exemption is not going to happen. (That’s why corporations have traditionally loved to incorporate in Delaware, of course–even many nonprofits.)
The only other likely regulating body would be the IRS, which could revoke its recognition of OpenAI as a tax-exempt entity under 501(c)(3). But the plot twist is that what would be a disaster for most nonprofit organizations could well be a big win for commercial interests, as investors such as Microsoft and NVIDIA might well push for a restructuring into a straight commercial entity—for example, converting into a Delaware Public Benefit Corporation (a move OpenAI has already signaled interest in). Of all the states in the United States, Delaware is the jurisdiction most likely to play ball with commercial corporate interests.
While I agree with most of these as terrifying possibilities, there is a very hopeful part of me that sees the consumers and appreciators of art as playing a huge part in preventing this. When everything is automatically generated and grows more and more generic, they will seek out books and other art from people simply because they’re not being satisfied with the crap they’re getting. Maybe I’m overly optimistic but, I can’t see it being the end of art as we know it. People will traffic in the real things, even if illicitly, and maybe even at a premium.
Art making is one of the earliest human impulses there is. It isn’t going away.
I think you’re right. The question is, how will the artist prove it’s not generated by AI, at least in part?
With any physical object, it’s trivially easy: just make a video of the act of creation. Probably this kind of video will be an important part of the provenance of the work going forward.
For digital works (including written works), it’s harder, but there are still sketches, notes, preliminary drafts, and so on.
There’s also quality. This whole conversation assumes that AI-generated works are indistinguishable from human creations. That simply isn’t true now, and I’m not convinced it will ever be true. There’s a good amount of evidence that suggests LLMs, at least, are reaching the point of diminishing returns. There’s also the issue of model collapse, and the fraction of human-created material in the training sets goes down.
I suppose you’re right, yet even this sort of thing can be faked well enough to fool a great many people. It’s rather depressing.
AI “art” will fool people who never look at art. AI “writing” will fool people who never read.
In any case, AI will continue to improve until it’s based on quantum computing, and then perhaps the only way to distinguish it from human-created art (including writing) will be that it consistently reaches the master level, surpassing what almost all humans can create.
This is an expectation that simply cannot be met. LLMs can’t and won’t ever surpass what humans can create because it’s missing the one thing that qualifies it as art, humanity. No matter how much existing work it scraps and steals, it does not have experiences, ideas, opinions, or beliefs. It simply cannot do this.
Put me on the list of people who are extremely skeptical about quantum computing, either by itself or as an augmentation to AI. Quantum computing is barely even where integrated circuits were in 1958, and the physics to make it work are fundamentally more difficult than for integrated circuits.
I’d like to be wrong. But as we can already mimic the neuron-based process of the human brain with rather limited hardware capabilities (a process not used in large language models, by the way), and since we’ve just now figured out how to build chips with multiples of quantum bits, it’s only a matter of time before we marry the mimicked brain function with the immense ability of quantum computing to explore all mimicked neural pathways simultaneously–something we humans cannot do.
You tell me. What does this portend?
I’m not so sure. I’ve just read an article where we’ve figured out how to add quantum bits to a chip where the qubits are at a distance, making them immensely more stable and practical. Even if we disregard the ability of qubits to explore all possibilities in a probabilistic manner free of linear time, let’s not discount that particle superposition gives us the ability to store information free of mere binary processing, as particle spin position can be zero to one, and in infinite range in between. Imagine the efficiency of processing in such a system. The power of LLMs will be nothing compared to where things are moving.
I don’t accept your premise. A number of very well-funded attempts to “mimic the processes of the human brain” have failed to model anything more complex than a dragonfly. There is no known solution to the enormous thermodynamic challenge involved in scaling quantum computers to a size that can do useful calculations, much less “explore all mimicked neural pathways simultaneously.” So I think you’re wrong, and even if you’re right, it won’t happen in my lifetime.
Yeah, I read that paper, too. Cool physics, very very far away from a commercially viable tool. (Here’s the link for anyone interested. https://arxiv.org/pdf/2403.12021)(At least I think this is the one you mean.)
Yep, that’s the one.
The paper describes how we’ve built a system that can trap and control over 6,000 atoms using laser “tweezers.” This is quite a leap because of the sheer number of qubits and that they maintain their quantum states long enough for use in computing. This is now scalable, folks.
Now, what exactly do you think companies like Microsoft and NVIDIA are doing when they invest billions in the likes of OpenAI? Is it to have the inside track on LLMs? Well, sure–but that’s short term. The real play is combining practical quantum computing with another type of AI that mimics the way our brains work–but at a level that no human brain can achieve.
It’s happening, folks.
But anyway, no, I wouldn’t like to see AI tools added to Scrivener. 