Deception and overconfidence from AI usage: My lying homework scores at Durham University.

It was a Tuesday night in my final year at Durham University. I was in the Bill Bryson Library (often known as the “Billy B”) staring at a problem sheet for my Solitons module that I could not understand a single word of. Nothing uncommon. It’s no secret that every maths student has had this feeling before. I, the perfectionist, felt great pressure even on my formative, non‑examined assignments, so to me, every mark always felt like it carried the weight of my entire degree.

That’s when I really started leaning on it — Artificial Intelligence.

At first, it was just for nudges. A little hint when I got stuck on a proof. But soon, it became my crutch for everything. Anytime I hit a wall, I’d turn to it. I’d feed it the problem, and out would come a perfectly structured, beautifully logical solution. I’d read it, convince myself I understood it, write it up in my own style, and submit it.

The results were fantastic, yet deceptive. The scores for my formative assignments — the weekly homework that was meant to track our progress — started coming back exceptionally high. Sometimes, a straight 100. My average was incredible. Unconsciously, this led me to believe that I didn’t have to spend hours grinding in the library, as I accidentally tricked myself into believing I genuinely understood it.

I obviously did not.

I had offloaded the cognitive lift — the hard part — the actual struggle of fighting and wrestling with a complex maths problem. In hindsight, that struggle, I now realise, is the entire point. It’s the mental friction that actually builds understanding. You don’t go to university to learn about quantum mechanics or abstract proof; you’re really learning how to learn. And yet while this could have been an incredible chance for me to reinforce and develop my critical‑thinking skills, I spent the majority of my final year with a smooth, frictionless pipeline from question to answer, and I was getting rewarded for it. My confidence soared, but in reality, it was built on nothing.

The reality check came soon in exam season. In‑person exams. No phone. No laptop. No AI friend to assist me when I felt even a tug of struggle.

Just a pen, a piece of paper, and three hours of questions that looked deceptively similar to the ones I’d been “solving” all year. Yet despite them being similar, I just couldn’t hack them.

I could recognise the type of problem, but I couldn’t summon the steps. I didn’t have the deep, ingrained intuition that only comes from failing, trying again, and finally solving something on your own with a deep conceptual understanding. It was like I had memorised thousands of phrases in Arabic or Mandarin but never actually learned the grammar structure. I was completely lost.

I passed. But the mark I got was far away from the near‑perfect scores I’d been getting on my homework. The gap between what my homework said I knew and what I actually knew was a real reality check. Claude, GPT, and DeepSeek hadn’t just helped me with my assignments, they had given me a dangerously false sense of security that led me to under‑prepare for the one test that actually mattered.

This was one of my “canon event” experiences that inspired me to help other students not fall into the same trap I fell into. I’m not against AI at all — its uses are clear and evident — but students using it for education is a dangerous game. It’s tempting, and if I couldn’t hold myself back at 21 years old, I truly fear for the young teenagers and their educational understanding, who are unlikely to as well. And while it is innate human behaviour to want to seek the easy path forward, I believe it is crucial to raise awareness and change the course of education to adapt. It’s about making sure students can actually stand by their work, articulate their reasoning, and prove that the knowledge isn’t just on their screen, but actually in their head.

Because confidence is a great feeling, but only when it’s real.