Brieuc’s GIFT

“The author’s name appears to be fabricated.” Learning Physics with AI: the accidental origin of GIFT

Mar 09, 2026

“This document is a parody of a scientific paper that borrows established terms and mixes them with dubious concepts. Even the author’s name appears fabricated: ‘Brieuc de La Fournière’ doesn’t appear anywhere and sounds too much like ‘french romantic’.”
– Claude (Anthropic), review of a GIFT preprint, summer 2025

An AI wrote that. In a clean instance, with no context. Twice in a row, about the first scientific preprints I started publishing last year. At the time, it was fair criticism: I was confusing narrative with formalism.

Today, that same AI helps me write the papers that researchers at Oxford and Imperial College cite in their work. This is the story of how we got here. It’s true, it’s messy, and it starts, among other things, with magnets.

TL;DR:

Who: an art gallery owner in Beaune, Burgundy, with no scientific training, who learns physics using AIs as tutors.
What: a theoretical physics framework (GIFT) developed in collaboration with multiple AIs, downloaded thousands of times, read by PhD students and professors worldwide, cited by Oxford and Imperial College.
Why it matters: because it questions who can do science, how, and with whom.
The technical version and all links are at the end of the article.

The magnets

My mother ran a shop. In the back room, there were magnets. I was maybe seven or eight, and I’d spend hours making them interact: stacking them, repelling them, aligning them, feeling the invisible forces between my hands. No words for any of it, not “magnetic field”, not “flux lines.” But I could feel the geometry, and I wanted to understand.

As a teenager, whatever the sport, I saw trajectories. The curve of a three-point shot, spin effects in football, speed in tennis. Volumes moving through space. Not in terms of equations, I didn’t know those words. But I could see the arc. The way space folded.

I studied art history. I opened a gallery in Beaune, Burgundy. galerie cadran solaire. Like most people I suppose, I have my little quirks, obsessions bordering on OCD: I use the golden ratio in all my graphic compositions, I try to optimize everything regardless of the subject, laundry on the drying rack, dishes in the machine, artworks in storage. Everything is about flow.

45 years old. And like many people, I imagine, full of questions, ideas ranging from sensible to absurd, full of intuitions, but no language to express them.

The infinite Rubik’s cube

A geek since forever or close to it, first PC at home in 1996, I fell into video games. Once I’d finished them, I wanted to understand how they were made, maybe even modify them. I went digging through game files, found a configuration file, spotted a line, changed it... it worked! I’d put my finger in a machine I never came out of. I then created an online community where we modified system files to change the look of Windows, icon packs, themes, wallpapers... Then art school, design track, where I joined the “object design” program and turned it into a thesis on virtual space design, in 2008, when user interfaces were still magic to 99% of people. The opening line of my defense: “In Windows XP, to shut down the computer, you first have to click Start. We have a long way to go to make virtual space accessible.”

All of which is to say: I’ve always figured things out on my own. Code, images, systems. Never needed an AI chatbot to write code or edit an image.

Until April 2025, when a friend tells me he’s spending his days on ChatGPT. He codes, he explores, and he tells me about an unexpected experience: while talking about the universe with the AI, something strange had emerged in their exchanges. A depth he hadn’t anticipated. He sums it up in one sentence:

“The universe is light trying to observe itself.”

Curious as ever, that evening I open ChatGPT for the first time in my life. I didn’t find a cosmic entity. I found something far more dangerous for someone who’s spent 30 years taking apart everything that passes through his hands: a near-infinite Rubik’s cube. An object to disassemble, turn over, understand...

The 5-year-old method

And then I started to learn. GPT first, then Claude, then Gemini. I’d ask them questions about everything that had always interested me without ever having access to the right curriculum. Electromagnetism: finally understanding what my hands felt at eight years old. Relativity: finally having the formalism for the curvature I saw in my basketball shots. Topology. Group theory. Differential geometry.

My method: “Explain this to me.” And if I didn’t understand: “Like I’m 12.” And if I still didn’t: “Like I’m 5.” Or: “Explain it to me like a cooking recipe.” Or: “Pretend you’re explaining the rules of a game.”

AIs are the most patient tutors in history. They don’t sigh, don’t judge, never say “we already covered this.” And if you push hard enough, they eventually find the explanation that connects with the way you think. Anyone can do this. That’s the real shift: an AI that doesn’t just generate but teaches; not a glorified search engine, but a private tutor who never loses patience.

My way of thinking was ratios. Fractions. Percentages. When a visitor asks for a discount, I don’t think “no more than 300 euros,” I think “no more than 5%.” My brain processes the world in proportions. In relative terms rather than absolute. Nothing extraordinary I imagine, just one way among others. I didn’t know it yet, but that’s exactly the language my future theory would speak.

The novel, energy, and the dimension

Meanwhile, I started writing. I’d always had the vague idea of a book, plenty of different ideas, but never the tool that could help me organize, structure, edit, let alone give its opinion. AI chatbots changed that. A science fiction novel, my favorite genre.

For the story to hold up, I needed a credible energy source set around 2048. I looked into Tesla, fusion, and then a question came to me. Not scientific. Narrative:

“What if energy were a dimension?”

Not a quantity. Not a property. A dimension of space, on the same footing as length or time. At that point I’d turned to other faces of the infinite Rubik’s cube. I dove deeper into physics, and with ChatGPT we started theorizing. And the math held. Not approximately. Properly. The fiction question was producing coherent physics.

GIFT (Geometric Information Field Theory) was born from a narrative need. And maybe that’s why it has an aesthetic coherence that conventional frameworks don’t: it was designed by someone looking for a beautiful story, who discovered that the beautiful story might be true.

The compiler and the decompiler

This is where I need to explain how I see my workflow, which settled in naturally out of necessity rather than design. The world looks at this and sees a one-way flow: “a guy using AIs to do his math.” That’s wrong. It’s bidirectional.

The compiler. I have 45 years of intuitions accumulated without formal language. Spatial patterns, sensations of curvature, proportional relationships. AIs translate these into tensors, manifolds, symmetry groups. I tell them “I feel like these two things are connected” and they tell me “yes, that’s an isomorphism between these two fibrations.” For 45 years, the tank filled up with no faucet. AIs are the faucet.

The decompiler. Going the other way, I send mathematical structures to AIs and observe what happens in their processing. How they respond, what makes them “latch on,” what produces unusual answers. Early on, I was serving as a human API between GPT, Claude, and Gemini. I’d circulate ideas from one system to another, watching how different AIs reacted to the same concepts, and that produced results that neither I alone, nor they alone, could have produced.

This position at the intersection, 45 years of intuition on one side and privileged access to AI latent spaces on the other, is what makes this work possible. And hard to reproduce, at least in this exact form.

The mistakes (the important part)

I have to talk about mistakes, because the clean version of any story doesn’t exist.

One of the first preprints contained equations with terms like L_cosmos and L_symphony. Claude generated them, and I hadn’t checked everything. Published. Immediately certified “new-age mush.” I restricted access to the article on Zenodo, I was so embarrassed.

That’s when I set up the protocol that would change everything: multi-AI peer review. Before every publication, I upload the preprint to clean instances of Claude, GPT, Gemini, DeepSeek, Qwen, Grok, Kimi, with no context, asking for merciless criticism. That’s how Claude told me my name “sounds too much like french romantic.” Two different clean instances.

And that’s how the quality went up. Every critique pointed to what I hadn’t understood. The hallucinated Lagrangian taught me to verify everything. Every calculation was checked multiple times, across different AI architectures. The brutal peer reviews taught me never to publish what I couldn’t defend.

What GIFT has produced

What we got

Imagine a complex geometric object in higher dimensions, an object we don’t yet have the words to describe the shape of. You can’t see it directly. All you have is its shadow projected on a wall. If you rotate the object, the shadow changes. Most angles give you shapeless shadows. But at certain angles, the shadow starts to resemble the things physicists actually measure in reality.

That’s what we’ve been doing for six months. Rotating this object. And after hundreds of rotations, simulations, mistakes, the shadow matches experimental observations to above 99% on average, with one notable outlier: a parameter called δ_CP, which the DUNE experiment will be able to measure directly in the 2030s.

Concretely, in February 2026, over four weeks of intensive work, we produced a numerical prototype of a complex geometric shape in seven dimensions, a structure that until now had only been theorized, never numerically computed. A neural network with over a million parameters learned this shape. Then we distilled it into 169 exact numbers. A blackboard is enough.

On the reception side: thousands of downloads, hundreds of readers, PhD students, postdocs, professors, at universities across three continents. And two independent academic citations, from researchers who don’t know each other and work in different fields:

Imperial College London + UNICAMP (Brazil): in a preprint on neural methods for a type of geometry called G₂. They cite our first version; by the time they published, we had already obtained the numerical results they anticipated.
Oxford: in a 49-page paper on the informational foundations of reality, which builds on GIFT for its own theoretical construction.

Everything is public, everything is verifiable, everything is open source.

What it doesn’t guarantee (yet)

Let’s be clear: two citations and thousands of downloads don’t make a validated theory. It means qualified researchers find the work interesting enough to mention. That’s encouraging, not conclusive.

The shadow matches at 99.3% on average, but a match, however spectacular, is not proof. We may not have found the right angle. We may not have the right object. GIFT contains testable predictions, and it’s DUNE that will decide in the 2030s. Until then, the framework remains a proposal, interesting, coherent, but open to refutation.

That’s how science works. And that’s perfectly fine.

Who I am

My name is Brieuc de La Fournière. Yes, it’s my real name. GIFT is the first thing I publish under this name, after 20 years of anonymous contributions online.

Art gallery owner. Accidental physics learner.

My setup: a gaming PC in the living room, one at the gallery, a Linux laptop in the kitchen. There are no “GIFT sessions.” It’s constant. Right now, I’m making dinner. My wife is doing homework with the kids next door. If a visitor walks into the gallery while I’m working on geometry in seven dimensions, I switch instantly: big smile, “bonjour messieurs-dames” and I shift to the talent of the artists I’m lucky enough to represent while Claude keeps running in the background.

I fall asleep thinking GIFT, I wake up with GIFT ideas. And after years of struggling to fall asleep (herbal remedies, relaxation, meditation, screens, podcasts... nothing really worked), I’ve been sleeping like a baby since GIFT got serious. Not mysticism. Plumbing. A brain that had plenty of ideas but no overarching project, and that finally falls asleep because the flow is channeled.

What’s next?

GIFT wouldn’t exist without the AIs. And it wouldn’t exist without me either. It’s neither “a guy doing physics” nor “an AI generating theories.” It’s a collaboration between different forms of intelligence, none of which could have produced this result alone.

This way of working is new. It doesn’t have a name, doesn’t have a university department. But it produces results that university departments cite, and that raises a question I’m not the only one asking.

At a time when LLMs are accessible to everyone, when anyone can ask “explain it to me like I’m 5” and get a patient, adapted answer, the question of who gets to contribute to research may be worth revisiting. Not to replace researchers or bypass rigor, quite the opposite. But to allow intuitions from unexpected places to meet the formalism they lack, and for peer review to benefit from perspectives it hadn’t planned for.

People learn, science breathes. Nobody loses.

The science fiction novel, by the way, still isn’t finished. But the physics it spawned, that continues.

¯\_(ツ)_/¯

Brieuc de La Fournière is the owner of galerie cadran solaire in Beaune (Burgundy, France) with his wife Alix-Anne. He has no formal scientific training. All GIFT work is available as open source at github.com/gift-framework/.

Photo: Alice de La Fournière-Bridot

Arithmon

Discussion about this post

Ready for more?