Last summer, I had the privilege to be a research intern at the National Center for Supercomputing Applications. I went in without any expectations, and with the mindset that it’s just an experience. I had been under the impression that computer science research is, just like any other research, all about reading papers and somehow writing more.
What I found out later was that a great deal of AI/ML research involves writing code and running lengthy and tedious experiments and thousands of API queries, the cost of which should be covered by your lab. In the process, I saved a lot of time by using Cursor. I felt very powerful, but also knew that it would be very embarrassing if I end up being caught by the PI (principal investigator) or research assistant. I don’t think using it to write code that conducts experiments is wrong, it’s just that I was worried it would hallucinate and fake results in the process, which it did.
Multiple times, instead of computing metrics of using an AI API, it would ‘simulate’ data because a prerequisite didn’t exist, so it was somehow a good idea to just fake data. This is why it’s important to always keep an eye on AI code agent output.
Currently, Cursor offers a free year of Cursor Pro subscription. It definitely has saved me a lot of time, and I think it’s time you try it out too.
hi
Comments (2)