AI Summary
The video critiques the feasibility of large-scale quantum computing, citing a new paper that calculates the physical resources required for a fault-tolerant quantum computer. The creator argues that the size, cooling requirements, and noise issues make practical quantum computers unlikely, and questions their real-world applications beyond research.
Chapters
The US Department of Energy, following an executive order, plans to deploy the world's first fault-tolerant, scientifically relevant quantum computer by 2028. The creator sarcastically suggests this is unrealistic.
A new paper is the first to calculate the physical resources needed for a large error-corrected quantum computer, focusing on superconducting circuits, the most advanced current approach.
The visible hardware is mostly cooling equipment for a tiny microchip. Current chips have at most a few thousand qubits, but a serious quantum computer would need millions, making the system enormous.
Assuming 10,000 qubits per cooling apparatus, a 4-million-qubit quantum computer would cover over 800 m², the size of a German Aldi supermarket, and double that with manipulation and readout equipment.
The creator argues that error correction becomes impossible at such scales, and alternative approaches like PsiQuantum's photonic qubits have not been demonstrated. Executives of quantum computing firms are selling shares.
In 2023, Wellcome Leap offered $5 million for a healthcare problem where a quantum computer beats a conventional one. No one won, indicating a lack of practical use cases.
A Los Alamos team found that conventional algorithms match or beat quantum versions on every benchmark up to 1,000 logical qubits, suggesting quantum AI may not offer advantages.
The creator admits cherry-picking bad news but notes that the only solid use case is breaking old ciphers, which doesn't justify the investment compared to other public benefits.
The creator remains highly skeptical of quantum computing's practical applications, arguing that the physical and noise challenges are insurmountable and that current investments are misallocated.
Mentioned in this Video
💡 Key Takeaways
First Physical Resource Calculation
This is the first attempt to quantify the resources needed for a large error-corrected quantum computer, providing concrete numbers for the debate.
00:41Size Comparison to a Supermarket
The vivid comparison to an Aldi supermarket makes the scale tangible and highlights the impracticality.
01:50Wellcome Leap Challenge Failure
A $5 million prize with no winner demonstrates the lack of practical quantum computing applications in healthcare.
03:27Quantum AI Loses to Classical
The finding that classical algorithms outperform quantum ones on benchmarks challenges the hype around quantum AI.
04:10Only Use Case: Cipher Cracking
The admission that the only solid use case is breaking old ciphers questions the massive investment in the field.
04:56Full Transcript
[00:00] Quantum computing has a deadline. Following an executive order of President Trump, the US-American Department of Energy has announced that they will “deploy the world's first fault-tolerant, scientifically relevant quantum computing capability for
[00:16] research and development by 2028.” Awesome! Can we please also have an executive order for quantum gravity by 2030, and while we’re at it, immortality by 2040? Yeah,
[00:29] I think it’s not going to happen, and today I want to tell you why. Stay around until the end for a special offer that will bring your coding skills to an entirely new level with boot.dev.
[00:41] For years I have told you that I think quantum computers are not going to scale and now the authors of a new paper finally put the numbers behind it. It’s the first attempt to calculate the physical resources that a large error-corrected quantum computer
[00:55] would require. They look specifically at superconducting circuits as qubits, which is the most studied and most advanced current approach. The problem with quantum computers is apparent if you just look at
[01:08] this footage. This thing which you see here has little to do with computing, it is mostly a noise buffer and equipment to cool a tiny microchip at the very bottom. These chips may contain 1,000 or maybe a few thousand of these qubits at most
[01:24] before they simply get too large for the cooling apparatus. If you want to build a serious quantum computer that can do all the things we’ve been promised, you’d need a few million qubits. Then the problem becomes you have to combine multiple
[01:38] of these cooling devices. Leaving aside that no one really knows how to do this without ruining the required entanglement, it becomes a problem simply because of the size.
[01:50] The authors of the new paper say, let’s look at something like 4 million qubits, which is a reasonable estimate, and be generous and say that we can put about 10,000 qubits in one cooling apparatus. The entire quantum computer would then cover an area of more than 800 m²;
[02:08] that’s about the size of a German Aldi supermarket. And that is without any of the necessary equipment to manipulate and read out the thing. Add all that equipment on top and we’re talking about at least twice the size, not counting all the extra facilities necessary to
[02:24] build the components. So if you were thinking we’d have a factory that produces quantum computers like laptops, it’s not going to happen. The factory will be the quantum computer, basically. And with a size like this the noise problem is going to explode. I am sorry,
[02:41] but really I see no way you will ever error correct anything that size. Now look, there are other approaches to quantum computing that use other technologies, like for example PsiQuantum that wants to use photons as qubits. But no one has even seen a
[02:55] demo of the thing. In what I am sure are completely unrelated news, top executives of leading publicly traded quantum computing firms have sold off significant numbers of their shares. It is not that I am somehow against quantum computers, I am just extremely sceptical
[03:11] of their supposed practical uses besides the research opportunity of just studying a big, highly entangled state. And keeping these researchers employed. That, to be fair, is a quantum computing application with excellent experimental support.
[03:27] Let me give a few timely examples to explain why I am sceptical. In 2023, the American non-profit Wellcome Leap offered $5 millions to anyone who would find a healthcare problem where a
[03:41] quantum computer could beat a conventional one. They published the results in April. The winner was… no one. No one could find a good use case for a quantum computer in healthcare.
[03:53] This isn’t an isolated case, I have previously reported that whenever someone looks closely at what a quantum computer can supposedly do the answer is not much. In a recent paper, a team from Los Alamos looked at the question of whether quantum computers are
[04:10] actually good to speed up neural networks, “quantum AI” as the headlines like to put it. They found that the conventional algorithm would match or beat the quantum version on every benchmark they tried, up to more than1,000logical qubits. So if you want
[04:26] to make quantum artificial intelligence more efficient, just remove the quantum computer. The authors say that it might be that there are some sorts of datasets for which quantum
[04:38] computers might have an advantage, but that hasn’t been demonstrated and then there is again the question of who would actually have a use for that. Likewise, salesman problem and related logical and logistic problems or finance applications.
[04:56] Now look. I openly admit that I am cherry-picking bad news to confirm my biases. But that doesn’t mean this bad news doesn’t exist. The one so far solid use case of quantum computers remains cracking old ciphers. That is not nothing,
[05:11] but I think it doesn’t remotely justify the investment. It pains me to think what good all this money could have done had it been invested in something that benefits the public. Like science communication that could have prevented this utter insanity in the first place.
[05:27] Artificial intelligence can do a lot, but if anything, it's increased the demand for software engineers who understand how to code. And there's no better place to get started than with today's sponsor, Boot.dev. Think about everything you've done today,
[05:42] and chances are most of it was accessing some sort of data through a user interface. Because that's how most apps work, how most businesses operate, and how most research is presented. And
[05:55] that's exactly what you'll learn on boot.dev, backend development, database management, user interfaces. They have hands-on lessons in which you write code directly in the browser
[06:07] and you solve challenges to level up. And it's not single scattered lessons. They have very carefully designed learning paths that'll get you started on anything from Python for beginners, Typescript
[06:20] to your first personal project. If you decide to go for the full experience, use my code SABINE to get 25% off your first year. So go and check this out. Thanks for watching. See you tomorrow.