AI was great until it was not

When I arrived at Oxford, I informed what many would call a traditional shock. Beyond the flamboyant parties and apparent micro-aggressions of the gowned elite, my discipline and math courses felt radically distinct than before. The lectures were considerably additional impersonal; the professors were a bit hostile; the assemblage was huge, and everyone had item to prove. This would have felt fairly normal if I had decided to continue my learning in Argentina, but, instead, I had adapted to the American learning model—at a academy akin Northwestern no less, which had a fairly small discipline program. Not lone that, I spent fractional of my undergraduate learning surrounded by PhD students, taking the identical classes as them.

But a masters in England was different. It felt bigger, additional ominous. Most group opted to study and resolve issue sets alone, and whenever we tried to activity in groups there was an anticipation to execute as the “genius” type. I had PhD offers from Harvard and Oxford direct from undergrad, and I felt akin I had to demonstrate I deserved it. It was partially my fault. I did awareness the need to brag in command to be taken seriously, but it backfired on me. It was clumsy at first, but the surroundings shifted drastically mid Michaelmas term. Pretty shortly our discussions and ego games shifted from meaningless whiteboard debates to my classmates merely stating “I fair asked ChatGPT how to do it.”

ChatGPT?

This was in delayed 2024. I had heard of AI in the past. Everyone was raving concerning it all year, but no one I knew had used it for class. I saw group complaining concerning how much h2o it used, how awful it was for the surroundings and a few group equal stated that it power slay us all someday. This felt akin complete nonsense to me. I recall whenever I was penning my undergraduate thesis in mid-spring. I opened instagram to procrastinate for a few minutes, and I saw a famous “PhD influencer” (whatever that means) say she put her name in ChatGPT, and it ruined a lot of data concerning her and how awesome she was. I idea it was a bit scary for a “website” to drag up all your data akin that (and I established the endeavour a bit narcissistic), but I was deeply curious, so I gave the chatbot a chance. I recall asking several inquiry concerning scattering amplitudes, and it replied alongside nonsense. So, I pressed it and asked again alongside additional detail, and it said:

“You should ask a theoretical physicist.”

And at that point, that was adequate for me to near the opening and never open it again.

That is, until a few months later, I was studying at Oxford alongside my “friends” at the time, and they kept raving concerning how ChatGPT could do their homework. Problem sets for most of the MMathPhys courses at Oxford are not graded. Real grades arrive from exams or your dissertation. A few classes necessitate you to present homework, but ‘passing’ is enough. I idea it was a bit meaningless to ask a chatbot to do everything for you whenever the complete intent of the homework was to assist as exam practice. Yet, in January 2025, I established myself stuck trying to resolve one QFT question, so I asked the chatbot akin my allies recommended, and, again, it stated several bullshit. I didn’t appearance at it again until a few months later.

It was a bit odd how the idea of “homework” had seemed to decay. Before, at least, we had the decency to appearance at Stack Overflow, and scout for akin problems or issue solving techniques that could help, but now abruptly group weren’t equal doing that. They fair asked ChatGPT to do everything. It defeated the intent of doing homework to be honest. Though, I think we had misplaced that a during ago. I continually idea that beyond practicing for the exam, issue sets were awesome to study what collaboration during issue solving is like.

But in reality, collaboration has rarely been part of my actuality as a student. The final period I recall really collaborating alongside chap students was in my freshman twelvemonth of college. I was taking multivariable calculus alongside a bunch of engineering students. I knew two girls, Andrea and Juliana, from inclination and we immediately cast a small collection conversation to activity together. We would try all issue set separately first, and afterward all of us would dispatch our answers to the collection conversation and we would debate. For instance, one female would ask me “oh, that’s a clever way to do this question. How did you cognize we had to merge complete this curve? Can you explain it to me?” or inform me “Hey, I think this inquiry is wrong! You should use Lagrange multipliers this way instead.” And I would do the identical item for them. I learned a lot in that class, and I partially due it to my awesome collaborators at the time. Debating, discussing and fair learning from them was genuinely lovely. It helped me rotate into improved at math and really grasp all the concepts.

I was deeply enthusiastic to commencement the discipline important and collaborate alongside another physics peers. In the autumn term of my sophomore year, I signed up for an honors differential equation math category alongside a companion who was additionally a discipline major. As I idea was customary, we sent all another our solutions to check. It was my archetypal period taking an honors math class, during that companion had taken three before. I additionally had to young female the archetypal two lectures since my scholarly adviser insisted I wasn’t prepared for honors, and I spent a week on a deeply boring norm category before I convinced the prof to let me switch. That is to say, I made a ton of mistakes in that archetypal homework. So I messaged my companion akin I had done in former and started asking: “Oh, it seems I misunderstood this concept, what do you average here? Why is this true? I missed this class, so I idea we had to do item different” to which she accused me of trying to plagiarize her activity and started ignoring me completely, and telling another group I was trying to copy her. Our companionship was never the identical following that. And I learned a difficult lesson: I was alone. For the remainder of the course, I worked on homework alone, I studied for exams alone, and I ended up alongside an A in the class. Actually, I spent most of my flat operating alone, equal as an undergrad among PhD students. I would explain my answers to my classmates, but I rarely had that rear and onward from freshman twelvemonth again.

That being said, ChatGPT fair made the surroundings surrounding collaboration additional bleak. Now, there was nothing beyond “Just ask ChatGPT.”

I was motionless hesitant to use AI. I saw many group in the MMathPhys use it for not lone their homework, but for their dissertations and the investigation they expected to publish. I was confused. Another companion started using it as a therapist. I was motionless deeply confused. I was repelled by how agreeable it was, but it did provision the weird illusion of having person rearward the screen.

Regardless, in May of 2025, all my bad decisions caught up to me. I really spent most of my period at Oxford doing nothing. I continually considered attending lectures several kind of social ritual. They have never helped me. I have continually learned from books. Similarly, several issue sets were dense, uninsightful and useless, and if they were not needed for my grade, I wasn’t bothered to do them. I additionally had a PhD recommendation waiting for me at Harvard, so there was really nothing to lose. I alternatively traveled a lot, went to parties, embarrassed myself, peruse a ton of books, played chess, fed the ducks and cried a lot. But at several point, I realized I had to get my masters flat and I had a duration and a fractional to study four Oxford MSc classes from scratch and continue my exams. I was cooked.

And there was ChatGPT. The item concerning ChatGPT is that, well, it’s not a person. It doesn’t “think” akin we do. But I realized it was extremely helpful for one thing: rearranging what humans stated in a additional digestible way. So, I would sit alongside my cosmology and broad relativity notes, and all period I got stuck or I couldn’t visualize a definition, I would fair screenshot or copy the small division or declaration I did not comprehend and ChatGPT would re-explain it to me. Sometimes it included a quick example that helped me navigate indicator notation, for instance, or it would explain to me that torsion-free fair meant that parallelograms closed.

One of the chief incompatibilities alongside norm physics/math learning and the way I think concerning problems is that definitions and derivations in a vacuum are a bit useless to me. I can prosecute complex math and I can comprehend what a definition says, but it won’t click for me until I see a genuine explanatory example. This is not fair a me thing. This is how Feynman idea concerning discipline as well1, and I am certain a lot of group portion this cognitive profile.

But alternatively of dealing alongside peers that got annoyed at me and my stupid questions, ChatGPT provided that extend nearly immediately and alongside no fuss. After three weeks of using it, I realized I didn’t need it as much anymore. I could prosecute advanced concepts—like Penrose diagrams or Kerr Black Holes—on my own, create my own explanatory diagrams and derivations without AI. I took my exams two weeks following that. I learned the satisfied General Relativity I&II, Cosmology and Geophysical Fluid Dynamics in a duration and a fractional from scratch, during additionally taking the two Standard Model courses and a QFT in curved spacetimes class. My grades were between elevated passes and merits, but it was enough! I got my degree. And these were completely closed booked exams, so I did the activity myself. But ChatGPT really helped me accelerate my cognition acquisition.

That being said, I did not rely it at all for new research. Once I arrived to Harvard, I was stated my archetypal PhD project in first November of my G1 year, which is a bit first although not rare for elevated energy. I was operating alongside a postdoc, and I was pleasantly amazed to see that the skills I learned in undergrad from my two mentors were incredibly helpful for that problem. I embedded several twistor theory and chiral algebras into a nice mathematica notebook, which helped alongside a lot of calculations. I barely used AI on that project. It helped me discover backdrop writings and construe several paragraphs/derivations in those papers. Beyond that, I lone used Claude Opus to place one part of code in my archive to assistance me agreement alongside poles and residues in an expression since I wasn’t certain how to do it. That was it.

However, the sole minus document really changed the expectations surrounding AI in physics. Now, an inner example of OpenAI had established item researchers tried to activity out for months before. That was cool, but it had nuance. A lot of the dense lifting of that document and the conceptual leaps were done by genuine people, but the accomplishment of AI was not to be understated. It was impressive. The evidence was notable and not straightforward, and this document opened a lot of investigation directions in Celestial Holography. AI is really awesome at proving things inside a stated framework. It’s additionally fine at things that prosecute a strict logic or providing complex mathematical structures that are sound, but that’s not all that discipline is. Or, I akin to accept so.

The matter is that all period I showed my small mathematica notebook that I have coded mostly myself looking at the Wolfram documentation, the two my seniors and PhD peers fair would ask:

“Can’t ChatGPT do that?”

Maybe. But coding a issue in mathematica is not fair concerning “efficiency” or penning accelerated code. It is concerning encoding a theory, a framework, operating alongside border cases, and learning. Maybe I could have asked Claude or ChatGPT to create a chiral algebras mathematica library, and it power have been fine or improved than mine, but afterward what would be the point? Maybe it would have been faster, slightly faster. Maybe I could have fed it all the distinct things I wanted to do, and let it code on the backdrop and spit a ton of answers… but as a archetypal twelvemonth pupil that seemed to defeat the complete intent of doing investigation in the archetypal place. Furthermore, ChatGPT is not infallible. It can be very wrong. And not noticing whenever it’s incorrect is fair embarrassing for everyone involved.

I don’t accept in the grind of doing a difficult integral or evidence from scratch to demonstrate you are man adequate to grip it (if a calculator or mathematica can do it, why suffer?), but I do think there is item precious is solving an engaging problem, understanding a structure, and doing most of the conceptual dense lifting. I completed my archetypal project alongside extremely constricted AI use, and the document was published.

But whenever it was period to commencement my second project as a caller G2, I did notice the surroundings had changed drastically. If you stroll throughout the Harvard discipline division and lean into the offices, a lot of group volition have ChatGPT, Codex, Claude or akin open in their monitors. A lot of the emails I started to obtain contained AI generated notes or notes according to ideas stated by ChatGPT. It wasn’t fair me, a lot of my peers go through this too, current and everywhere. This is a average cognition now. Now, “brainstorming” method talking to ChatGPT. A lot of my period is spent deciphering ChatGPT notes, realizing it’s nonsense, and moving on to the next.

I was annoyed. I had objections, questions, and concerns, but I was close downward and told to ask ChatGPT. So, I asked ChatGPT and it fair “validated” everything I said, but didn’t propose any solutions. The issue is that I wanted to conversation alongside my superiors, ask questions… and I couldn’t. I realized a lot of doing a PhD in theoretical discipline is dealing alongside noisy group screaming at a blackboard, so it doesn’t awareness extremely interactive either. Some group say a lot of fancy words. Bullshitting is average custom several say. I realized at several item that it wasn’t fair me; there was several posturing, several performance, and ChatGPT made that worse. So, I was solitary alongside a chatbot again. But, now, I didn’t have a set of address notes written by a top physicist at Cambridge concerning a theory that was established nearly a hundred years ago… I had a brace of papers, vague ideas, and nowhere to go since on a PhD you activity out things no one has done before.

And it wasn’t good. Because I am fair a second twelvemonth PhD pupil and I don’t cognize anything. I cognize a lot of things, but in the expansive scheme of things, I am inexperienced and I need mentors. And a chatbot cannot be my mentor. And afterward I realized that identical tool that helped me so much a twelvemonth ago was drowning me and my field. I don’t think technologies are evil, but I do think how humans utilize several technologies can be harmful. But it’s not AI’s fault, it’s ours.

I did acknowledge first on that a lot of group in my site were small focused on finding the truth than they were on projecting, being correct and showing off. I realized first on that for a lot of physicists instruction and mentorship is additional of a burden than item they deem particularly valuable. We desire the fastest PhD students, the ones that published three document in undergrad, the ones that complete a calculation the fastest. We desire the hottest topics and the most citations. But if a chatbot can substitute a PhD student, if it’s faster, additional productive and it can assimilate Weinberg’s QFT volumes in five minutes, why do we need students at all?

Why do we collaborate whenever we can fair activity solitary alongside a bot?

Remember whenever I mentioned I had that collection conversation alongside my engineering allies and we were continually comparing and brainstorming? Sometimes, they would capture a idea or see a spotless resolution I didn’t notice, and I would do that for them too. That helped us grow, and it gave me a profound understanding of a lot of those concepts. I would resolve exams in unexpected ways, but that was fun! It helped really grasp the math and how beautiful it was. But that was additionally a fit collaboration. It wasn’t concerning getting the highest grade, or having the smartest solution, gatekeeping or screaming. We would sit down, try it out, politely discuss, and explain. We wouldn’t be average or condescending, we would fair study from all another and enhance all other.

People keep us in check. Each individual is different. We all have a distinct background, a distinct way of seeing the world. We arrive from distinct training. But, you study from others and they study from you too.

And we are losing that now too. Should we fear that colleagues operating in AI companies power obtain our prompts and try to get onward to resolve a issue before we do? Should we fear for threats? Should we fear that our days as PhD students volition be spent telling a device to “try harder” alternatively of deeply understanding item new?

Truth is, I was extremely impressed and enthusiastic concerning how ChatGPT helped me study. Would I fairly have receptive professors, lecturers that cognize pedagogy, physicists that desire to compose understandable document alternatively of purposefully burying understanding in meaningless jargon, and peers I can conversation to without having an anxiety spike? Yes. I would fairly have that one hundred percent. And I fear that now, it’s all going to be worse. Because we now have in our hands a device that validates all of our biases and buries them in dozens of pages of meaningless math.

I am fair a PhD pupil who feels a small defeated, not since AI power obtain my job, I am privileged adequate for that to not be a involvement (and let’s be honest, I am privileged adequate to have the chance to do a PhD in theoretical physics). But since the way we are using AI is destroying what made discipline fun in the archetypal place. Is it really fun to publish thirty document in a week using AI fair for the purpose of it? One of my allies in the program rightfully asks “Are those document value publishing?”

The complete item of solving an engaging issue or understanding the mysteries of the cosmos is not to win an award or have the most citations, but to get nearer to that nearly mythical item we call truth, to cognize the cosmos and ourselves through the study of the nature of things.

The issue is that the small chatbot in our pockets is making us additional secluded and in so, it’s making us stray additional from this humble goal: to learn. But it wasn’t fair the chatbot, it was the civilization about academia too. I think chatbots are a mirror, a deeply validating sycophantic mirror alongside the cognition of the internet. But mirrors can be hazardous if we remain for too lengthy too charmed by our own reflection, and we ignore concerning everything and everyone alternatively too. In particular, it would be beautiful sad to demolish among the few things that makes our existence meaningful, not the fancy award or the fancy position, but merely having a community to portion and conversation and learn. To study things. That is a beautiful thing.

-Nia