Asked about Controls, confounds and pre-registration
What is a confound, and how do you control for it?
An answer that works
Something that changed alongside the variable you meant to test, such as extra compute or longer training. Control it by matching the baseline on everything except the one change, including parameter count, data and training time.
Reading an answer is not being able to give one. The section that teaches this is Controls, confounds and pre-registration, 4 hours of Experiment design.
Learn this properlyAlso asked about Experiment design
How do you know your improvement is real?What is an ablation study and why does it matter?How many seeds should you run?Why decide your metric before running an experiment?Your new method beats the baseline by 0.8 points on one run. How do you decide if it is real?Your method has three components. How do you show which ones matter?