Bitter Lesson on LLM Alignment Research from an Undergrad

During my final year of undergraduate, I spent most of my college time on doing research. Luckily, I got into two research internship programs, both from MBZUAI with different mentors. The first project is about uncertainty quantification on Large Language Models (LLMs), while the other is about improvements in Direct Preference Optimization (DPO) by utilizing uncertainty.

Both projects have a simple idea to test, but the cost it takes to verify that idea is very expensive. Tens of hours of compute using RTX PRO 6000 is required and in the end it gives uncertain results. I’m still questioning whether it’s the right topic to pursue because of it. Should research be this expensive, or am I choosing the wrong topic to dive into?