During my PhD, I’ve mentored a bunch of undergrads — some later went to CMU, UIUC, Cornell, UW etc. But honestly, most of them only ever touched one small part of the research lifecycle. They never got the full end-to-end experience of actually doing research.

Lately I am increasingly convinced that, with AI’s help, a motivated undergrad can actually do a mini research project all on their own.

So I found this undergrad from the same program i was in — literally 0 research experience. I told him: “Pick any topic you’re genuinely curious about. Let’s speedrun a workshop paper.”

He said: “I wanna build an AI that generates the best cheat sheets for exams.” And in my head I was like… “Bro that’s not research, that’s just an app.” But fine... interest matters. Maybe there’s something fun in it.

We started using our own AI-native research platform to brainstorm and review papers. I didn’t guide him much — I just watched how he interacted with the platform. At first, the AI kept spitting out these “fancy but useless” ideas. I was like 'Ok fine, next one please...' HOWEVER, after a second thought… I realized I was toooo stubborned like a old professor

That “boring” cheat sheet idea actually involved:

- limited pages → limited resources - knowledge format optimization → information density - picking which topics to include → importance, difficulty, frequency, score weight - objective → maximizing exam score

And the AI also pointed out: “this is a Knapsack Problem.” We even got the AI to run a quick experiment to validate the approach. Whole thing took maybe an hour.

I know it’s not any big breakthrough, but for a student’s first-ever project, it’s really cool

I was educated by AI again this time: Science often starts from simple curiosity — not from grand theories. The best research happens when you try to solve real problems and accidentally uncover general principles along the way.