“For aspiring AI researchers, working on mechanistic interpretability is likely a better career focus than working on new model architectures due to it being a less crowded and more fertile area.”