Comment by SirensOfTitan

What is the probability that newer models are just overfitting various benchmarks? A lot of these newer models seem to underperform GPT-4 in most of my daily queries, but I'm obviously swimming in the world of anecdata.