Reflection on First (Likely) Publication

May 24, 2025 Reflection

Recently, I had the chance to work on my first research project in machine learning, specifically AI safety. I went in with a lot of assumptions about what research would be like, and many of them turned out to be wrong.

Although I’ve been reading and discussing many of the latest papers on AI safety, this was my first time working on true research, or so I thought. I had many expectations. Many of them incorrect; others, blatant misconceptions. This is an attempt to lay them out, see why I had these expectations in the first place, and how I can learn from them.

Let me start with my biggest misconception: Science is a structured and organised method of getting closer to the truth.

Over my childhood years, the notion of Science developed into something like an ideal, standardized method, together with everyone associated. Although I later learnt that there are other ways of doing science than, “Propose a hypothesis, design an experiment, and test the hypothesis”, I didn’t believe it. Thus, I entered this project with the expectation of doing Science à la Karl Popper. Now I’ve seen that certainly isn’t always the case, and most likely (at least in machine learning) isn’t how many scientists conduct their research.

Rather, it went something along the following lines. Everyone involved was busier with other stuff. There was little coordination. No one knew what others were doing or had done. Almost all work got done in the last week of the 2 months. New proposals of what should be in the main paper were done days before the deadline. Stress, stress, and more stress. Sleep deprivation, sleep deprivation, and more sleep deprivation.

But, there were also so many unexpected positive things that happened. So much exhilarating uncertainty, so much laughter, the beauty of seeing how people can rise up in moments of extreme stress and urgency, the thrill of working under stress, and, most importantly, advancing Science, however little it may be.

My misconception of Science didn’t stop at the method itself, but as mentioned before, also influenced my notion of the people associated with it. Since Science was the supposedly ideal method of getting closer to the truth, scientists had to embody all its principles. This turned out to be equally incorrect. I thought I had internalized the idea that scientists are also just people trying things out when I read my first nonsense paper (and the many following it). I thought I knew that scientists were merely trying their hardest, and not always succeeding in following Popper’s Path. Turns out, I didn’t fully believe it yet.

I joined the project and thought that the work would get delegated, everyone would put on their metaphorical labcoat, and follow the Principles of Science. That wasn’t the case. Everyone just tried their best, and given the many limitations of us humans, that can often result in less than what we hope for.

Also, I have realized doing research under so much stress is, except for the thrill of successfully submitting the complete research paper 5 minutes before the official deadline, unenjoyable. To me, Science is the pursuit of truth. A process that should not be disturbed by inconveniences of deadlines and the accompanying stress. Rather, it should be a careless activity, something which one can truly emerge into, and subsequently rise up from a bit closer to the Truth.

Many thanks to the people allowing me to join this project. I learned a ton from it, and, besides the sleepless nights, wouldn’t have had it any other way.