A new quantum paper has proven how easily the media is fooled.
Let’s say you have two lists of numbers (we’re talking computers, so it’s how a language model represents “dog” and “cat” internally). And you want to know how similar they are. Do you see a cat, or a dog?
Classical computers multiply corresponding entries, add them up. Done. Takes microseconds.
This paper did something complicated instead. It took those same two lists and used the numbers to configure the starting state of a quantum bit. A standard quantum operation was used (the Hadamard gate) that causes the two values to interfere with each other like waves. It measured the result many times, like sitting on the beach watching the waves roll in. The statistics of measurements tell you roughly how similar the original numbers were.
This technique has been known since the 1990s. It’s in textbooks (Nielsen & Chuang, Quantum Computation and Quantum Information, 2010, Chapter 4). The paper just confirmed that if your two lists of numbers happen to come from a language model instead of anywhere else, the technique still works.
Yawn.
It’s like demonstrating that a calculator can add numbers, and the numbers came from a spreadsheet. The calculator doesn’t care where the numbers came from. The fact that text originated in an LLM is immaterial to the computation.
The “quantum” framing and “semantic interference” language make it sound like quantum mechanics is revealing something about meaning.
It isn’t.
The quantum circuit is just a much more convoluted way to multiply and add, which is all cosine similarity ever was.
The paper appeared in what looks like a predatory journal (generic name, rapid acceptance, author self-cites two prior papers from the same journal, affiliation is “Financial Physics Lab, Finland” which isn’t a recognized institution). That context explains how something this empty got published, and why outlets like The Quantum Insider ran with it uncritically.
Check the source. There are much more novel and interesting uses of waves.