Timothy Nguyen

Quantum field theory textbooks have been lying to you or have left you confused. For a typical passage like the following from Peskin&Schroeder - what does it mean to do a change of variables on an ill-defined path integral? For perturbative QFT, my paper resolves this issue: 🧵

@adilsoubki @pangram I asked GPT to summarize the paper. Took over an hour of retrying and then post-generation revising. I’d say that’s quite meaningful effort on my part.

@IAmTimNguyen @pangram slop?

“Deep learning is alchemy” may be the most repeated criticism in AI. It also misses the mark. Alchemy failed to deliver results. Deep learning, by contrast, has produced transformative technologies. And fields like medicine are only partially understood without being deemed alchemical. So calling AI “alchemy” captures part of the problem, but not all of it. Modern AI is not simply undisciplined experimentation. It contains significant amounts of rigor. But we still struggle to answer basic questions: • Do models understand? • Why do they generalize? • When will they fail? The deeper issue is that rigor takes different forms—and in AI, those forms are unevenly developed. My new paper distinguishes three: • Conceptual rigor: coherent terminology and paradigms • Epistemic rigor: reliable scientific understanding • Operational rigor: reliable performance and deployment This framework helps explain both the extraordinary progress of modern AI and the uncertainty surrounding it. Conceptual rigor asks whether the field knows what it's talking about. • What exactly is intelligence? • What qualifies as AGI? • What does it mean for a system to be aligned? Consider the debate over whether current models are intelligent. One person points to their breadth of performance. Another points to weak planning. Another emphasizes sample inefficiency. Another asks whether it has a grounded model of the world. They appear to disagree about one property. Often, they are evaluating four. This is why conceptual clarity matters in practice. Questions about intelligence, understanding, AGI, and alignment do not remain confined to philosophy: they shape how things are measured, optimized, and built. Epistemic rigor asks whether empirical success has become scientific understanding. The paper focuses on three criteria: • Can findings be reproduced? • Can behavior be predicted in advance? • Can success and failure be explained? AI experiments are unusually reproducible in principle: code, data, and models can be copied. But conclusions may still depend heavily on random seeds, hyperparameters, implementation choices, benchmark selection, and compute budgets. Reproducing a number is not always the same as reproducing the conclusion drawn from it. Prediction is harder. Scaling laws can forecast some training outcomes. Infinite-width theory can lead to more tractable settings. Classical learning theory explains important pieces. But we still lack broad principles telling us when a model will generalize, fail under distribution shift, or remain robust under adversarial perturbations. Explanation is harder still. Neural networks are mathematically specified, yet their learned features resist human interpretation. A behavior may arise from training data, optimization dynamics, internal representations, or interactions among all of them. The system is transparent in code but opaque in meaning. Operational rigor is where modern AI is strongest: benchmarks, evaluations, monitoring, red-teaming, and deployment controls. The field has become highly effective at improving systems without first obtaining a scientific theory of them. Benchmarks turn capabilities into measurable targets. Post-training shapes behavior. Tools and scaffolding compensate for model weaknesses. Operational rigor can therefore partially substitute for scientific understanding. That imbalance defines the deep-learning era: • Capabilities rise rapidly. • Explanations lag behind. • Benchmarks become optimization targets. • New systems generate new phenomena faster than theory can absorb them. AI is advancing while continually changing the object that science must explain. For AI to mature as both a science and a technology, it will require all three forms of rigor: • Clearer concepts to define our goals. • Stronger science to predict and explain system behavior. • Better engineering to make systems genuinely reliable. The future of AI depends not simply on demanding “more rigor,” but on identifying which kind is missing—and understanding how the imbalance shapes what we can build, know, and control.

Paper link: arxiv.org/abs/2607.03634 I was asked to contribute a chapter to Routledge Series on Philosophy of Rigor on the topic of AI. So here it is, my first paper in philosophy (of science)! Hope I did alright.

@luchuubit I think it’s that a smaller community enables trust. And highly intelligent and ambitious people, who are well supported, usually have high integrity. Caltech only has around 900 undergrads.

@IAmTimNguyen Honest question - what is stopping Caltech's honor code from being adopted elsewhere? Cultural reasons, or simply because student body elsewhere tends to be much bigger & unpoliceable using the student community alone?

Back when I was an undergrad at Caltech, all assignments and exams (including the final) were take-home, timed, and self-administered. The Caltech honor code was real and lived. Looking at the impact of its alumni, that trust clearly paid off. I'm not sure if that model is still feasible in today's age of super-human LLMs, but one thing hasn't changed: the ultimate test set, where the only final score truly matters, is the real world after graduation.

@ShiqianMa @arxiv Finally got the hold removed 2 days ago. Can't help but wonder if social media posting provided the bump. (I'll take the win!)

@IAmTimNguyen @arxiv i submitted one on 5/19. Already 6.5 weeks. Still on hold.

How common are papers posted to the @arxiv (in cs/AI) held in the hold queue ridiculously long? Currently have a submission on hold for over 5 weeks. I get that the arxiv is understaffed and gets a high volume of submissions but surely they can do better. Some form of triage involving LLMs to detect submission quality and heuristics on qualifications, institutional reputation of the author(s), and whether the paper is already accepted for publication (true in my case) would probably go a long way. At this point the length of the hold does more damage than it does good.

@_____Lightning In an ideal world, the cheaters would be exposed at the interview stage or on the job.

@IAmTimNguyen All of which you said sounds great in theory, but all or most jobs are based off of gpa, so if the people who cheated have higher gpas, the opportunity is not given to those that didn't cheat.

@REasther Yes. (My memory was foggy and was erring on the affirmative, but Google search confirms it's open book.) The problems were sufficiently hard that the book won't save you if you don't know your stuff. And even if you had a bit more time if you were not rigorous about your timing.

@IAmTimNguyen Were they intended to the solved open-book? (There is a fairly finite set of viable physics exam questions, IMO, so it is not hard to see how LLMs get to be particularly good at them given that corpus of material provided by obliging academics)

@ChrisInterno @klindt_david @arxiv @ChristianInte16 Yes the service desk has been unhelpful - generic apologies that there is no update on the length of there hold.

@klindt_david @IAmTimNguyen @arxiv @ChristianInte16 Yep, can confirm. It's usually a mix-up with the primary category. we'd picked AI at first when it should've been CV. Just reach out through arXiv's support portal (arxiv-org.atlassian.net/servicedesk/cu…) that's what fixed it for us :)

@ShiqianMa @arxiv You up I’m the same boat.

@Yi_Zen_Chu @arxiv Oh wow. That is unfathomable.

@IAmTimNguyen @arxiv Notice it made sure to state that publication in a journal does not guarantee publication on the arXiv. Yes, paper is currently under review; but I am worried about freedom of expression and inquiry in physics. The arXiv has become a de facto monopoly on physics publishing.

@Yi_Zen_Chu @arxiv Is it under review somewhere? Come back and see if arxiv will honor hosting the paper once it’s accepted?

@IAmTimNguyen @arxiv The full content of their objection is in the screenshot. I fear there's not even enough substance to judge. OTOH, I need to get my paper peer-reviewed in order to be posted on the **pre-print** arXiv? x.com/Yi_Zen_Chu/sta…

@Yi_Zen_Chu @arxiv Ouch. Is there merit to their assessment?

@IAmTimNguyen @arxiv Rewrote the Kirchhoff diffraction integral in position spacetime (as opposed to the usual frequency space); discovered a generalized Poisson-Arago spot. Posted it to physics.optics. @arxiv took 2 weeks to send us back a terse response to reject the paper.

@star_stufff @arxiv Unsure that’s why I’m asking. This is my 2nd time for a long hold and both times was cs.AI

@IAmTimNguyen @arxiv Wait, is this specifically true for CS or for other domains as well?

Poindexter Puzzle No. 10: Ethos, Scheme, Feline, Robert, Urban, Facet, Canon, Drown, Influx, ??? What adjective of noble character comes next in this sequence? Submit your solutions in this form before July 9th (all correct entries are entered in our monthly prize draw): forms.gle/KsUCZB9ppHmLXc…

The job market is crazy these days.

@KempeLab @amilabs @ylecun Exciting Julia! Hope to see great things coming out of the lab.

I am excited to share that I am joining @amilabs as Director of Research, Paris, working with @ylecun and an exceptional founding team. Further progress in machine intelligence will require not only scaling foundation models and large-scale engineering, but also new ideas and breakthroughs. This is what makes AmiLabs such a unique and exciting place to build. Its focus on world modeling — systems that can learn richer representations of the real world, reason, plan, and learn from interaction — is a long-term research direction I am deeply excited about. I am particularly looking forward to working with extraordinary co-founders @sainingxie, @lxbrun, @michaelrabbat, @pascalefung, @mavenlin, @laurentsolly, and the amazingly talented AmiLabs team, to helping build the research organization and to the journey ahead.

@nolightupstairs I don't know, but Hilbert say, being a formalist might fall into that camp instead of the Platonist one.

@IAmTimNguyen Are there a significant number of mathematicians who would claim that math is invented?

What's your take on math+AI?

My thoughts: If you're a mathematician, might your reaction to the AI encroachment be based on your answer to that age-old question: Is mathematics discovered or invented? The most straightforward, even if oversimplified, two-way division: You're a Platonist, you see math as discovered. For you, you enjoy savoring mathematical gems from the mountain-top, to see the peaks and valleys that existed prior to your arrival. You would enjoy basking in as many vistas as you could. Or you regard math as invented. For you, climbing the mountain cannot be excluded. Being transported to the summit defeats your sense of accomplishment and struggle. The journey rather than the destination is the primary value. How do these perspectives align with your attitude towards math+AI?