Quantum Won’t “Save” AI
I've seen an uptick in commentary and headlines along the lines of, "Oh well, current large language model progress is plateauing, so we won't have Artificial General Intelligence next month; but with quantum computing, we'll soon have it, because... quantum" (waves hands).
I've worked in the quantum computing sector and am still in contact with my former colleagues (👋 shoutout to the 1QBit team!) to say with reasonable confidence: Quantum computing won't do anything meaningful for the sort of AI a business would care about, and certainly not for large language models / generative AI, for the foreseeable future.
Yes, important and exciting work is happening. Progress is steady. Multiple players are advancing the state of the art, and I'm certain that great things will come of that.
No, none of this matters for AI systems that work at the scale of a GPT-5.
Quantum computing is not a drop-in replacement for classical computing, where you just replace a conventional CPU with a quantum processing unit (QPU) and off you go. Instead, it's specialized hardware designed to solve incredibly narrowly defined problems, such as factoring a large number or determining the ground-state energy of a spin glass. The latter is what the D-Wave quantum annealing hardware is designed to do. If you do some clever math, you may be able to cast other problems you actually care about in those terms, particularly in scheduling and optimization. None of these use cases matters for training a gigantic machine learning model. (There is a quantum algorithm for solving linear equations, but its requirements in terms of the number of qubits are beyond ridiculous for current quantum hardware.)
In a way, the computational paradigms behind AI and quantum are opposed to each other; on the AI side, we're dealing with staggeringly large models with billions of parameters, on the quantum side, we're (currently) dealing with, at best, dozens of usable qubits.
It's almost as if, now that the irrationally exuberant hype is wearing off, certain tech influencers (and CEOs of quantum hardware companies?) latch onto the next topic for their hype. Blockchain. VR. AI. Quantum. All of these have kernels of usefulness that are at risk of being crowded out by undifferentiated hype.
Instead of dreaming about living in the Star Trek universe with sentient androids, holodecks, and faster-than-light travel, let's focus on solving actual problems with existing and proven solutions.
