Quantum Traction Theory
Notebook

I Asked an AI to Roast My Own Physics Paper

The brief: put on a classical physicist’s glasses, land cold on my Kerr paper, and say exactly how bizarre it looks. Reader, it found a cathedral built to house a doorknob — and it was right.

I do something for fun that I cannot fully recommend: I keep more than one AI assistant around, the way some people keep more than one knife. One of them is endlessly patient, unfailingly polite, and writes like it is drafting a peace treaty between two countries that quite like each other (you can probably guess which). The other one — Claude — is funnier, and has roughly the patience of a senior professor at 4:55 on a Friday afternoon. When I want a warm second opinion, I ask the diplomat. When I want the truth with the skin left on, I ask Claude.

This week I wanted the truth with the skin left on.

So I gave it one instruction, more or less this:

And then it became that professor. Gloriously. Completely. With a body count.

It opened friendly, which made what followed worse. It recognized the spine of my paper on sight — Δn = λ K E², the Kerr effect, John Kerr, 1875, drawn on a thousand blackboards — and admitted its guard went down for about four lines. Then, in its words, it found itself “surrounded by words I have never seen in an optics paper.” Artian geometry. Two-clock projections. An absolute background clock. Access rails. You could feel the eyebrow going up.

It got worse from there, and I mean that as the highest compliment. A sampling of the damage:

• On my carefully-derived trace-free tensor: correct, it allowed — but “why summon three axioms with cosmological-sounding names to obtain a fact elementary representation theory hands you for free?” It called my supporting apparatus ceremony. On reflection, I have been extremely ceremonious.
• On my argument that the effect must go as E²: it informed me, very dryly, that “finite does not imply quadratic.” Which is true. Parity does that work in one line. I had dressed up a one-line symmetry argument in a much longer coat and hoped nobody would check the pockets.
• On the cos(π/8) ≈ 0.924 sitting at the front of my Kerr constant: “Why 8? Why not 6, or 12? It is a half of a half of a right angle. I have no hook to hang it on.”
• On my finest formal flourish — an entire chain-complex apparatus, boundary maps and a metric and a trace, wrapped around one optical coefficient — it noticed I had inserted square roots specifically so that squaring them through the trace would hand me back the square I started with. Circular gift-wrapping. Then it delivered the sentence that became the title of this post.

A cathedral built to house a doorknob.

I laughed out loud. Alone. In the Starbucks near La Défense. At a sentence a machine wrote about my life’s work. This is, apparently, who I am now.

It was not finished. It reached my two-clock structure and the angle 7π/48 — which, yes, is just π/8 + π/48, eighths and forty-eighths of π — and it wrote, in the margin of my soul, a single word:

Numerology?

And it gently observed that an “absolute background clock,” the thing my entire time sector hangs from, is Newtonian absolute time wearing a false moustache — which is the one idea relativity spent a century throwing out the door. The relativist in the professor was not amused. The relativist in me has heard this before, usually in the mirror.

The verdict it landed on is the part I cannot stop thinking about. It called the paper, in a word, inverted: everything in it that is rigorous is standard physics it already teaches, and everything genuinely new is, so far, either invisible — I admit cos(π/8) can be absorbed into an unknown and never caught — or unexplained, a number from a book. The working parts are not new. The new parts are not yet doing measurable work. The exotic machinery, it noted, sits exactly where I concede it cannot be tested.

Here is why I am grinning instead of sulking. That was my take too.

I did not write this paper because I cannot see how it looks to a classical eye. I wrote it because I share that eye. I find the vocabulary strange. I find an absolute clock suspicious. I wince when small fractions of π land near measured numbers. The whole reason to build a skeptical professor and hand him the chalk is that a theory worth anything should survive being read by its harshest imaginary referee — and the fastest way to find your soft spots is to let someone, even someone made of arithmetic, press on them in public.

And, to its credit, the professor was fair where it counts. It noticed the dimensional analysis is correct. It noticed that I explicitly refuse to fit anything to the measured Kerr value. It noticed that the paper says, out loud, that it has not yet derived water’s actual Kerr number, and names that as unfinished business. Those, it conceded, are the moves of someone trying to be honest rather than someone selling something. I will take that.

Best of all, it named the exact experiment that would make it stop smirking: take two liquids — water and nitrobenzene — predict the ratio of their Kerr constants from independent molecular data alone, touch no measured Kerr value, and see whether the universal cos(π/8) cancels and the numbers land. If they do, the strange machinery has finally earned a number. If they do not, the whole thing falsifies itself cleanly — which, the professor admitted, is more than most fringe papers ever dare to set up.

So that was my week. I asked a language model to roast me in the voice of a man who would never answer my emails, and it handed me a title, a to-do list, and a thoroughly deserved bruise.

I am keeping the professor. He is cheaper than a referee, he never sleeps, and he has not once told me what I wanted to hear. Next, I am going to build him the water-and-nitrobenzene ratio he asked for, and we will find out together whether he eats his hat.

He does not have a hat. He has glasses. I will build him a hat first.