Recent AI model progress feels mostly like bullshit

I first used GPT 2.5 many years ago through a google colab notebook. These newer models aren’t all that much better. I’m sure they do better on their tests because they’re trained to beat them but really the biggest “innovation” was a UI and API.