🧵 3. Despite advancements in AI training methods, including model scaling and human feedback, the study reveals that larger models struggle with accuracy on simpler tasks.