I skipped CVPR for an interferometer in a box

Trip 1 to 5 June 2026 · Written up 5 June 2026 · Last expanded 29 September 2026

I landed in North Carolina on a few hours of sleep, which is not the state you want to be in for a week of quantum mechanics. It was my first time on the central and eastern side of the state, with the whole Research Triangle in front of me. Those dates had been earmarked for CVPR in Denver, right up until word reached me that NC State was running a Quantum ML summer workshop, at which point Denver stopped being the plan.

The audience skewed hard in one direction, as far as I could tell: almost everyone was arriving from physics and heading toward computing, and I was going the other way. That is the position I wanted to be in, because I had just picked up some minor sustained cloud compute, and the week doubled as a live test of my planning platform (ProjectFlow, if you want the name). The question was simple enough, namely whether a tool can keep a relative outsider abreast of the technical detail coming at him in someone else’s field, in real time, for a week.

Transducing between vocabularies

This is the use of LLMs I keep coming back to, more than the drafting or the code completion: the reduction of friction between disciplines, taking an idea stated in one field’s vocabulary and its notions and transducing it into another’s. If that works, interdisciplinary collaboration gets cheaper and productive research output goes up as a side effect, since in my experience most of the barrier between two fields is not conceptual difficulty but that both sides have built separate names for things that turn out to be the same thing.

The workshop handed me an instance of exactly that, and it did not come from a machine. Before I had sat through a single talk, a fellow Georgia Tech alum in attendance gave me a heads-up on the work of Robert Huang, an apparent quantum advantage found in classical data processing, published only a few weeks earlier, i.e. a person in the room did the routing I had come here to test software on. I do not think that undercuts the thesis, it just sets the bar.

What I cannot post yet

I captured all of the QML workshop, and then found out how much of it I have to sit on, because a good portion of what was presented is still not in public prints. This has been running as an interesting talks series across three venues this summer, and NC State is the one where I have to be careful, so the most interesting material of the week is also the material I am least able to show you.

That gap is not new and it is not really about secrecy, it is the ordinary lag between the moment a result becomes known to a room and the moment it becomes citable to everyone else (the old scientific societies ran on that lag, sealed letters and priority disputes and all). What is different now is that the room is small, the write-up is instant, and the queue in front of publication has not shortened to match.

An interferometer in a box

The thing I have thought about most since is not a talk at all. At NCSU I saw an interferometer and photonics setup in a box, built as an aid for instruction in a quantum computing course, a whole photonics setup packaged so that a student meets the physics directly instead of reading about it. NC State and the Duke photonics lab over in Durham were the first two stops of the intellectual journey I have been taking Brutus on this summer (yes, the Ohio State one).

Fields turn into disciplines when their apparatus gets cheap enough to teach with. Microscopes and telescopes were bespoke instruments long before they were things you handed to a sixteen year old, and the handing over is what produced the next cohort of people who could use them, so a quantum optics kit that fits in a box is the same move, one layer up. I want one for Buckeye STEM students, and the same goes for Georgia Tech!

If a semester of quantum optics now fits in a box you can carry into a classroom, I do wonder how much of the rest of the curriculum is still built around apparatus we no longer need.

Who I met

• A fellow Georgia Tech alum in attendance, Quantum Machine Learning Workshop, NC State. Gave me a heads-up on Robert Huang’s newly published result within hours of my landing, before any of the talks. The human routing layer got there first, which is worth remembering.
• Robert Huang, author of the result being passed around the workshop. Did not meet him, but his paper on an apparent quantum advantage found in classical data processing, out only a few weeks before, was the first thing anyone put in front of me here.

Photographs

Links

Robert Huang’s paper on quantum advantage in classical data processing, and the ProjectFlow public board with the pre-populated assets for these talks. To be added.

Open entry
This page is not finished, and is not meant to be. I add photographs, people and links as I dig them out, so it keeps growing after publication. Why the whole site works this way is on the colophon.